Pages

May 10, 2013

The Geography of Hate

UPDATE (5/13/13 @ 10:45pm): We have written and published a FAQ to respond to some of the questions and concerns raised in the comments here and elsewhere. Please review our comments there before commenting or emailing.

Following the 2012 US Presidential election, we created a map of tweets that referred to President Obama using a variety of racist slurs. In the wake of that map, we received a number of criticisms - some constructive, others not - about how we were measuring what we determined to be racist sentiments. In that work, we showed that the states with the highest relative amount of racist content referencing President Obama - Mississippi and Alabama - were notable not only for being starkly anti-Obama in their voting patterns, but also for their problematic histories of racism. That is, even a fairly crude and cursory analysis can show how contemporary expressions of racism on social media can be tied to any number of contextual factors which explain their persistence.

The prominence of debates around online bullying and the censorship of hate speech prompted us to examine how social media has become an important conduit for hate speech, and how particular terminology used to degrade a given minority group is expressed geographically. As we’ve documented in a variety of cases, the virtual spaces of social media are intensely tied to particular socio-spatial contexts in the offline world, and as this work shows, the geography of online hate speech is no different.

Rather than focusing just on hate directed towards a single individual at a single point in time, we wanted to analyze a broader swath of discriminatory speech in social media, including the usage of racist, homophobic and ableist slurs.

Using DOLLY to search for all geotagged tweets in North America between June 2012 and April 2013, we discovered 41,306 tweets containing the word ‘nigger’, 95,123 referenced ‘homo’, among other terms. In order to address one of the earlier criticisms of our map of racism directed at Obama, students at Humboldt State manually read and coded the sentiment of each tweet to determine if the given word was used in a positive, negative or neutral manner. This allowed us to avoid using any algorithmic sentiment analysis or natural language processing, as many algorithms would have simply classified a tweet as ‘negative’ when the word was used in a neutral or positive way. For example the phrase ‘dyke’, while often negative when referring to an individual person, was also used in positive ways (e.g. “dykes on bikes #SFPride”). The students were able to discern which were negative, neutral, or positive. Only those tweets used in an explicitly negative way are included in the map.

Tweets negatively referring to "Dyke"
All together, the students determined over 150,000 geotagged tweets with a hateful slur to be negative. Hateful tweets were aggregated to the county level and then normalized by the total number of tweets in each county. This then shows a comparison of places with disproportionately high amounts of a particular hate word relative to all tweeting activity. For example, Orange County, California has the highest absolute number of tweets mentioning many of the slurs, but because of its significant overall Twitter activity, such hateful tweets are less prominent and therefore do not appear as prominently on our map. So when viewing the map at a broad scale, it’s best not to be covered with the blue smog of hate, as even the lower end of the scale includes the presence of hateful tweeting activity.

Even when normalized, many of the slurs included in our analysis display little meaningful spatial distribution. For example, tweets referencing ‘nigger’ 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. 

Hotspots for "wetback" Tweets
Perhaps the most interesting concentration comes for references to ‘wetback’, a slur meant to degrade Latino immigrants to the US by tying them to ‘illegal’ immigration. Ultimately, this term is used most in different areas of Texas, showing the state’s centrality to debates about immigration in the US. But the areas with significant concentrations aren’t necessarily that close to the border, and neither do other border states who feature prominently in debates about immigration contain significant concentrations.

Ultimately, some of the slurs included in our analysis might not have particularly revealing spatial distributions. But, unfortunately, they show the significant persistence of hatred in the United States and the ways that the open platforms of social media have been adopted and appropriated to allow for these ideas to be propagated.

Funding for this map was provided by the University Research and Creative Activities Fellowship at HSU. Geography students Amelia Egle, Miles Ross and Matthew Eiben at Humboldt State University coded tweets and created this map.

The full interactive map is available here: http://users.humboldt.edu/mstephens/hate/hate_map.html

113 comments:

  1. If I may ask, why weren't misogynist/sexist tweets included in this map?

    ReplyDelete
    Replies
    1. That's my immediate puzzlement too.

      Delete
    2. I would love to include misogynist tweets in this map (such as the word "bitch" or "slut"), but the pure volume of tweets including these words would not allow for manual sentiment analysis. As there are well over 5 million tweets using the word "bitch" and not all in a negative context, this would cost over $50,000 and take students more than 2 years of full-time tweet reading.

      Delete
    3. Could you explain how 'bitch' could be used in a positive context, aside from the general swearing 'son of a bitch'?

      Delete
    4. Some women, particularly young, relatively radical feminists, are reclaiming words like bitch, slut, and cunt. For example, the feminist quarterly magazine "Bitch."

      Delete
    5. Also, "slutwalk" etc, which people may disagree with, but it is an attempt by people to reclaim the word as part of their sexuality and it would be wrong to characterise it as a slur.

      Bitch, of course, could also be used to refer to a female dog. People do still do that occasionally.

      Delete
    6. I posted this below but I should have put it here. I'm involved in the dog world. Breeders regularly use the word "bitch" to refer to a female dog - as in "We lost our beautiful bitch when she was hit by a car." Also, I live in West Hollywood. Gay men here often playfully refer to each other as "bitch" - "Jake, stop being such a bitch." So I don't blame Humboldt for excluding this word - that's a lot of non-hostile tweets to sort out.

      Delete
    7. This comment has been removed by the author.

      Delete
    8. If a man calls a woman a bitch, which is rare, it is to spare her the indignity we inflict on men, which is to call them an asshole.

      Delete
    9. If someone pulls a prank on me, or scares the shit out of me, I usually come back saying "You little bitch!"
      Not really meaning to insult them. Any word that is originally an insult can be positive, despite it's intended use. Like the word "nigga" for example.
      "He's my nigga" to display companionship or friendship. It's a bit insensitive, yes, but slang is slang.

      Delete
  2. I'm curious to know if you included alternate spellings or misspellings in your search. There's more than one way to misspell the six-letter word used to defame gay men, for example.

    Also, why only "cripple" for ableist slurs? What about "retard" or "gimp"?

    ReplyDelete
    Replies
    1. Gimp was also considered but most of the tweets referred to the software. There were very few (less than 100) used as a derogatory term towards disabled persons.

      Delete
    2. Did you try and look at "retard" too? There is an entire movement speaking out about the word. Check out #Rword conversations on twitter or the handle @EndtheWord. Thanks!

      Delete
  3. This is the first iteration of a larger project, other terms/topics will be posted as they are processed.

    ReplyDelete
    Replies
    1. That's good to know. I'd be interested in knowing if "retard" is one of the terms your team is or will be processing. If so, have you considered "derp" as well, since that's rapidly becoming a new "acceptable" insult for the mentally disabled?

      Delete
    2. I'm really surprised racist language targeting whites was not in this iteration. That would have been interesting to see given the demographics of Twitter. It also might imply (to some) that whites are ones spewing the racist language on Twitter.

      Delete
    3. Seeing as white people can't experience racism in a society founded upon white supremacy, I think it makes sense to not include "whitey" among the ranks of "nigger," Rachel.

      Delete
    4. Yeah, I'm pretty sure whites aren't the ones living with the legacy of Jim Crow and lynching. Whites aren't the ones the GOP is *even now this very week* saying are mentally superior because eugenics and skull sizes and we're still in the 19th century or something. People upset because someone mentioned their white skin really need to get a reality check about systemic bias.

      Now, if we want to talk about the way that classist slurs can be targeted at subgroups of white people (trailer trash etc) then there's an interesting conversation to be had. Lily white folk whining that only the negroes get to complain about being oppressed, well, that ain't so interesting.

      Delete
    5. Rachel, white people are the ones spewing the racism on twitter. Read the FAQ's, it explains that there were not enough racial slurs against whites to show up on the map, making them irrelevant, regardless of how much you feel the institutionalized privilege of black people has held you back in Alabama.

      Delete
    6. O Suzi that was *nice.* ♡

      Delete
    7. in the immortal words of Dan Akroyd: Suzi you ignorant slut. (and i mean that in the best possible way)

      there is no scientific way to tie twits to race, or gender, or age... only quantity. the "racism" could be coming from martians with twitter accounts for all we know.

      Delete
  4. Hi, how are you doing the normalization to compute the frequencies? The main map looks simliar to the overall density of Twitter messages in the USA, so I was wondering if it had been normalized.

    ReplyDelete
    Replies
    1. I agree... I'd love to know if/how the map was normalized relative to population density.

      Delete
  5. This is great, but it mostly shows highly populated areas. Are you controlling for number twitter users in some way? You will see that there are lots of racists in an area, but that's just because there are more people.

    ReplyDelete
  6. Normalization is based on hateful tweets / total tweets.

    It clearly is not a map of population density. See California, lots of people but relatively few hateful tweets when compared to the total volume of tweets.

    ReplyDelete
  7. This isn't very scientific. It lists pejoratives spoken by only one group. Since it's impossible that only one group can be racist, and does not use epitaphs used for every group, this project isn't something you could really take seriously.

    ReplyDelete
    Replies
    1. What do you mean? Can you be more specific?

      Delete
    2. It is using hate directed from one group to others. There is nothing from other groups, there is also nothing about women ( which is a problem, unfortunately).

      To be a true "Hate Map", it would include all "hate", would it not? Otherwise, it just is hate from one group and not scientific. If you want to be proper, you must include every group, due to the fact the US is a nation of many ethnic groups. Using the Census as a model for the groups would have been simple and made more sense.

      All these people did, was create a heat map, which can be dismissed.

      My suggestion is to use the Census and try again. Then it can be irrefutable. It might take more time, but, that's what needs to be done.


      Thanks

      Delete
    3. The Census doesnt track tweets, or bigoted slurs. Yes, this site could post the Census, but I dont see what that accomplishes, or how its relevant.

      Heat maps can be dismissed? Anything can be dismissed i guess, i dont see how a bar graph or a topographical stylized map is "less dismissible"

      I know what you doing, youre not fooling anyone. You're avoiding saying what you actually mean in attempt to appear not racist, which is not working. "Other groups" means non-whites. Youre wondering why the map doesnt show more racism against whites. Well the map DOES show all the racism against whites. Its essentially non-existent and completely irrelevant on the scale of racism. And since that doesnt line up with your world view that somehow a tiny demographic of many different kinds of people is persecuting a huge majority of the most powerful people on the planet (white americans) you just say "no thats stupid". Thats the world man, you can disbelieve it if you want, but your blind if you do.

      If they "use the Census" (how do you propose they do that? "Just, you know, USE it") and come up with exactly the same map, you'll just complain about how the Census isn't accurate.

      This is a map of bigoted tweets in a county divided by the total number of tweets to come from that county. Then turned into the notorious HEAT MAP. This map couldnt be more straightforward.

      My suggestion is YOU make the map you want. It might take a lot of time, but apparently that's what needs to be done. So do it.

      Full disclosure: Im white, lived in mostly white towns. I lived in Boston in Roxbury for a while, the only time in my life i felt like a minority, and because im white i didnt even really notice. I have been the "victim" of racism exactly once, with a young man calling me a cracker. Im pretty sure he was making a joke. Either way it didnt effect me because it never happens to me, and i dont understand what cracker even means. Take that for what you will.

      Edit: I was just explained what cracker means. Im now offended by it. But im over it. Maybe its all free shit i get for being white, who knows.

      Delete
    4. I think when he said "other groups", he may have been meaning, Asian's (Chinese, Japanese, Taiwanese, Korean, etc), Indians (Native Americans, East Indians, First Nations Etc), Jewish communities, Hispanics, Etc. Etc. As determined on some pages....

      Ethnographic division into races from Meyers Konversationslexikon of 1885-90 is listing:

      Caucasian races (Aryans, Hamites, Semites)
      Mongolian races (northern Mongolian, Chinese and Indo-Chinese, Japanese and Korean, Tibetan, Malayan, Polynesian, Maori, Micronesian, Eskimo, American Indian),
      Negroid races (African, Hottentots, Melanesian's/Papua, “Negrito”, Australian Aborigine, Dravidians, Sinhalese).

      To say one specific race on a map is more important than a complete map with many different races that get hated on just as much and as often is just not right. Granted, the (as stated in the form above, not my personal word choice and no racism meant) Negroids were enslaved for generations, but they were not the only ones in history that were enslaved in one way or another. Jewish-holocaust, and during that "Jewish Holocaust" MORE Ukrainians were killed and enslaved in concentration camps at the same time and they sure as hell don't get repayments for their lives and loses like the Jewish did, they rarely if ever get brought up. We have slavery today still, in many ways and fashions (Labor, sex slaves, the child sex industries, etc etc), with numerous different races being used, not just the Negroids.

      On another note, Any race that is not classed straight Caucasian, is allowed to make ANY racist remark/joke against another race, including Mongolians, Negroids, Caucasians, and yet if a Caucasian was to say just a JOKE about ANY other race OTHER than against their own Caucasian's, they are considered being racist (and even then, some people will call a Caucasian a racist for making a joke about their own "race". ALL of these races are just perpetrating and continuing the act of racism, it is not just one or two races causing it all. Hell, within their own groups and races, they hate on each other. I live in a town with a lot of First Nation's people, they have continually referred to us as "Your people and our people" when we do not refer to them in the same manner, then they go as far as to hate on their own band members and family and have racism against their own. Human is human, it's not them or us against the other, we all need to stand together and stop hating when their is no reason to hate to begin with, There is no ONE superior race, and even if there was a "Superior race" I can definitely say it is not the Caucasian "race", or white as some say.

      Delete
  8. I second the suggestion to check for alternate spellings. Also, some who use offensive language online will use alternate spellings to avoid certain filters common on websites. Typing "fa99ot" or "ni99er"for example, could become a habit that would transfer to tweeting.

    ReplyDelete
  9. Why have you omitted derogatory words directed against Whites such as: honkey, peckerwood, white trash, white boy and cracker?

    ReplyDelete
    Replies
    1. Because they are racist against White people.

      Delete
    2. My guess--I am not related the project in any way--is because the power balance still dramatically bends towards white males. Any cursory look at a board roster of a Fortune 500 company--or our Congress--will tell you that. Sure, you can be "racist" against those in power--but it definitely does not carry the same threat or dominance play that "hating on" the marginalized does.

      Delete
    3. I couldn't find any references to searches for "hillbilly, hick, redneck, racist, goober, peckerwood " or any of the other terms commonly applied to Southerners, so this map is not a real reflection of all hate speech. But I will guarantee that I will have this beaten on my head because I am from the South. Institutionalized hate like the comic strip "Snuffy Smith" and various "redneck" reality shows constantly demean us rural types but are deemed "okay" and "harmless fun" because they are aimed at whites, yet even that distinction fails when looking at some of other white, ethnic subsets.

      Delete
    4. Being called racist doesn't equate to hate speech, you jackoff.

      Delete
  10. Is it possible to download the 150,000 offensive tweets along with the geodata in a KMZ for our own analysis? I work for a software company that specializes in Big Data and I am intrigued by how the information would be displayed on a heatmap using a linear distribution as opposed to the percentile distribution you use here.

    ReplyDelete
    Replies
    1. I'd like to see a full context of ALL derogatory terms. Not just against blacks and latinos.
      I'm sure some PR firm has access to them somewhere.

      Delete
    2. You are a ridiculous racist FountainHead, and you dont do your research, which is why youre a racist. if you actually read the FAQ, you would realize they already addressed this.

      First off, if you really want to see ALL derogatory terms, it would include misogyny first, not white people, which is what i know you mean.

      Second, the words wigger, cracker, and honky were included in the search words. They didnt find enough references to necessitate mapping it. Meaning that kind of racism doesnt really exist, and is not relevant to the conversation.

      Third, this is directly from the FAQ that you did not read.

      "If you are a disgruntled white male who feels that the persistence of hatred towards minority groups is a license to complain about how discrimination against you is being ignored, just stop. You can refer to all of our previous commentary on this issue from November. Though we have typically refrained from deleting asinine comments to this effect - those who choose to make these comments do more to prove themselves to be fools than we ever could - we fully reserve the right to delete any and all comments we believe to be unnecessary."

      So that ^

      Delete
    3. Who would have thought that an Ayn Rand acolyte would start a derail based on 'white oppression?'

      I <3 you, Josef

      Delete
  11. This is fascinating, but I'm concerned that you might not have processed enough data to draw meaningful conclusions.

    You (appropriately!) normalize by number of tweets in that area to counter the effects of varying population density,
    which ought to leads to a relatively continuously varying heat map. This is the case when zoomed out, but when one zooms in it becomes clear that individual hot spots are disproportionately affecting the average. For example,
    if I look at the "racist" heat map at a high level (level 2) it looks like central Minnesota, from the Twin Cities westwards, is unusually racist. However, when I zoom in there's just one county (Hutchinson) in central MN that's quite hot.

    Frankly it might just even be one individual, and there are so few overall tweets in the area that the normalization results in him or her affecting how the entire state of Minnesota appears.

    Perhaps with more data a less "chunky" heat map could be produced. Or perhaps a different normalization algorithm would be more appropriate, that mitigates the effect of individual discontinuities affecting an entire region?

    ReplyDelete
    Replies
    1. I'd like to second Raph's worry and add to it. When zoomed in to the homophobic tweets, there are widely scattered, circular hotspots that are virtually all far from population centers. This suggests to me that what you're measuring is individuals (or small individual networks) that use a disproportionate amount of homophobic key words. This would be expected based on your normalization mechanism as I understand it: dividing the number of homophobic tweets by total tweets *by county*. The lower the number of tweets being normalized to, the greater the effect of individual negative tweets. If negative tweets are relatively rare in general, then the poisson distribution tells us that the distribution of negative-to-total-tweet ratios is going to be much spikier in areas where a very small population is being measured. As an example, starting from the assumption of equal, low odds of negative tweeting, the odds of just by chance getting 50% negative tweets in a county where only 4 people are tweeting is much higher than the odds of getting 50% negative tweets in a county where 10000 people are tweeting. As a consequence, your results are going to be confounded with the degree to which people are grouped together in your normalization process. Areas where (i) people are mostly living in urban areas, or (ii) counties are bigger and hold more people on average, are going to show fewer random spikes in homophobic tweets. The west is both more urbanized and has much bigger counties than the east, which may be partly why the west looks so much better in your maps. To take care of this confound, you need to normalize by equivalent amounts of people in each geographical area. One simple way to do this (though not without it's own problems because of rampant gerrymandering) would be to use congressional districts instead of counties. Please check out this issue and re-make your maps, if appropriate! I think this approach is really exciting and I'd like to see how it turns out without this unequal sampling confound.

      Delete
    2. ditto this entire post

      Delete
    3. Theres a clearly explained reason why this isnt possible. They can only map geocoded tweets, which represent 1-5% of tweets. This isnt a sample, its everything they could gather. Yes it would be great to have a more accurate and detailed map, but they did the best with what they had, and id be willing to be its fairly accurate. It doesnt really seem to single out any section of the country for anything, with exception of wetback, which makes sense because up North that term really has very little meaning, due to the lack of a border. Still means that the hotspot of wetback is hotter than the whole rest of the border, but it doesnt make sense to compare it to Maine or Washington. This is just a presentation of the data, you have to make your own conclusions.

      Delete
  12. I agree with Raph in that it seems there are a few problems with this analysis. You are assuming an unbiased distribution of people who geotag their tweets. Perhaps people who geotag are more likely to express racist sentiment? Normalizing hateful tweets per twitter subscribers should control for some effects of population density, but not all. Can you perform a geographically weighted regression, with pop density as an influencing variable?

    ReplyDelete
  13. I am a bit upset that there is no religious "hate" and where is the "black racists" this is really starting to smell a bit bad, the anti christian, anti conservative, pro-constitution, pro-gun, anti-NRA does not seem to be counted, why?

    ReplyDelete
    Replies
    1. Just...no.

      First of all, none of the so-hated groups you mention are inherent characteristics of a person -- religious belief and political affiliation are choices -- and so are not included. Second, none of the groups you list have any history of experiencing systematic oppression in the United States.

      Delete
    2. "I don't feel like your suggestion and apparent upset counts for much"

      Translation:
      "You don't matter" "Hatred directed against you is justified because you are entitled (whatever that means)".

      This is a perfect example of bigotry and racism right there and I believe it reflects the prejudice of those who created this project.

      Delete
    3. @Taylor "Second, none of the groups you list have any history of experiencing systematic oppression in the United States."

      Okay...? So, you're racist.

      @grb Is that something you are planning on proving at some point or are you counting on all the little 'floating sheep' of the left to agree with your lack of education and inability to pay attention? (don't count too long, we wouldn't want you to get stuck at, say, 3)

      (oops, I said "inability." guess that makes me an "ableist.")

      Delete
    4. Zoe's Law: in every conversation about racism, at least one entitled white guy whose life kinda sucks for no systemic or cultural reason (possibly he's just bad at business, or couldn't get into law because his grades sucked) will pop up with a comment about prejudice against whites (ie: he didn't get a job and is blaming affirmative action and not the fact that he is seriously flawed in some way) and why nobody cares about it.

      Informed citizens are best to adopt the following tactic: roll their eyes, ignore him, and move on.

      Delete
    5. Then the name needs to be changed for this map. Hate is inherit everywhere. There is nothing in the information which states, what Taylor or Zoe is claiming.

      If they are going to build a hate map, and it doesn't include every group from the Census ( at the least), and ignores women, then this is just nonsense. People have a right to attack this as one-sided. You don't have to create separate maps for "Hate", because it can simply be homogenized.

      If either of you are the author of this map, you have failed. Please pick another subject and , hopefully, not anything related to science, thanks.

      Delete
  14. HEAT MAPs are probably the wrong visualization choice for this data. It is misleading since hate generally appears to be driven by urban vs rural living.

    A better mapping might be color filled counties, with no overlapping bleed, similar to voting district maps. Other possibilities include target circles.

    The current map makes it look like the entire midwest/east coast is hate driven, yet when you zoom in ANYWHERE, you see that the hate driving low population counties had bled over into the major acceptance cities like Chicago/Nashville/etc.

    Hate is often fueled by fear and lack of understanding of other cultures. In smaller communities, you are less likely to have friends of each culture so you are less likely to feel for them. In those small communities, the minority hides or moves away because of a lack of community support, increasing the problem. With lack of support, it only takes 1 person to spread hate and seed it in the community.

    An appropriate graphic illustration should highlight the local areas so that it becomes apparent what drives the hate.

    ReplyDelete
  15. I would also be curious if the map looks different if you "grouped" all tweets to individuals. In other words, instead of normalizing around the number of tweets, determine if individuals send any hate tweets, and then normalize around the individual tweeters.

    You used an example that 31 individuals sent 41 hate tweets. This could lead to isolated hot spots from angry individuals.

    Often you see an individual angry Facebook user who can't seem to send a single response without hate. There may be 100 benign comments from other users, but 30 hate comments were all from the same user, who was particularly active that day.

    This approach would isolate the "percent of the twitter population" that sends hate, vs the "percent of hate" among the tweets.

    Both would be meaningful in a media setting.

    ReplyDelete
    Replies
    1. Making the data granular on individuals strikes me as the best next step, both because Twitter already does this for you and because it disentangles the questions "Where are the racist people?" and "Where are racists most voluble?" There might be a correlation between not liking black people and not being able to shut up about it that could skew this representation.

      Delete
    2. I'd actually be interested in looking at this as a solution as well...would be interesting to topic model by tweeter to ID diffs between topics from "racist" folks and "not-racist" folks. I'd suspect that you might come up with some interesting differences.

      But I like this idea.

      Delete
  16. Hatred of women doesn't count? This is incredible. You build a complex application to map hate in the United States, and you leave out the single largest category: misogyny. Frankly the lacuna just reinforces the generalized misogyny of our culture, that hating women is so routine you don't even think to track it. For those of us who are women and the constant lifelong objects of that hatred, however, it's stunning.

    ReplyDelete
    Replies
    1. I agree with your point, but an not sure what specific words they would look for to "calculate" those statistics.

      Delete
    2. I would recommend words such as slut, whore, ho, c**t, b***h, etc. There are many words out there that are derogatory towards women. The fact that people can't think of these terms offhand just goes to show how much misogyny is engrained in our culture.

      Delete
    3. I agree, misogynistic language is as ingrained now as racist language was up to the 50s. This is evidenced often by the way disputes over other matters end up with such abuse - as currently in the secularist/atheist movement in the US. I think also more work needs to go into alternative spellings (and misspellings - some of these people are not very bright after all!) which is a much more complex task.

      I'm not convinced by what looks like an almost perfect E-W split. I think the results are too much influenced by population density.

      Delete
    4. Yeah, right. Don't forget to include words, like woman, girl, lady or ma'am. With all of the communist "feminism" B.S., these days, a female will slap you in a heartbeat if you show her any amount of respect.

      Also, since being a "slut, whore, ho, c**t, b***h," escort, hooker, "dancer," etc., is the new "cool" for girls, they should be left off the map altogether.

      Delete
    5. I like how you equate paternalistic behaviour singling out women in creepy ways with respect, and then make a bitter and baseless (sexist) statement about a younger generation that you likely have zero contact with, because they think you're creepy.

      Delete
    6. Despite your assumptions, I am confident the woman who made this map is well aware of sexism (particularly if you bother to review her other research). There are unique challenges to misogynist sentiment analysis as most anti-woman slurs are much more widely used in non-hateful ways, making it more difficult to suss out genuinely hateful tweets.

      Or you could assume that the female author was deliberately ignorant of this and chose to perpetuate sexism because she thinks it's dandy.

      Delete
    7. "I am confident the woman who made this map is well aware of sexism (particularly if you bother to review her other research)."

      There appears to be one woman on the otherwise male-dominated team of professors, and I see no evidence that the group has ever taken sexism seriously as a form of hate. I'm sure the one woman on the team is aware of sexism, but being aware of something is not the same as thinking it's important; women are strongly socialized to NOT think sexism is important, and to regard all other forms of discrimination as vastly more significant.

      If the authors were aware of how serious an omission it was to exclude misogny, why did the original post presenting the research make no statement about it? Only now, with the additional posts, do they explain that because people use the word "bitch" all the time it's simply not possible to track misogyny. As if "bitch" were the only slur that could be tracked.

      Delete
    8. @Sophie,
      I don't see anything wrong with what you're asking for. However, I think the volume of data for just race hate is overwhelming. It may be a good idea to do a separate search on gender discrimination. Not to include it in this one.

      Delete
    9. "It may be a good idea to do a separate search on gender discrimination."

      I agree, it sounds as if that would be necessary. But all the more reason that it should have received at least some acknowledgment in the original post.

      Imagine if the shoe were on the other foot, and the authors had presented a "Geography of Hate" map that contained zero reference to racism. None. Just categories for homophobia and ableism and, let's imagine, misogyny. No reference to racism at all in this "Geography of Hate" in America. It would be surprising, but you would read the accompanying post expecting to see some explanation of the approach, some reference at least to racism and the authors' decision to leave it out of the study. But imagine that the accompanying post had no such explanation. No reference to racism at all. Imagine that the authors just confidently talked about how the map shows "the significant persistence of hatred in the United States" despite showing nothing about a most notorious form of hatred that enslaved a large percentage of the population and subsequently subjected that same population to apartheid, terrorism, ethnic cleansing, and denial of human rights. It would be weird! You would be shocked. You would wonder why the authors didn't at least mention it, if at least to say "of course this map doesn't cover one huge horrible form of bigotry because it was just too big, so we focused on smaller phenomena we could track, like ableism."

      I don't mean to beat an expired horse, as it were; I'm trying to explain---for those who aren't accustomed to thinking of misogyny as important and terrible and dangerous---why the original post with this map was so bizarrely inappropriate.

      Delete
    10. @Sophie: Thanks for your concerns. They're duly noted.

      Delete
  17. Did you find any hateful tweets toward people of Middle Eastern descent? Or would it have been too difficult to distinguish racial hatred from religious/ideological hatred?

    ReplyDelete
  18. Is this raw data available in some way? Just the raw text of the tweets, with their geo-data and offensive/non-offensive tagging?

    If there's some way to anonymize the author, this information would be nice to have too.

    ReplyDelete
    Replies
    1. Even without the geocoding, this could be great sentiment analysis data.

      Delete
    2. Unfortunately that's not possible. The one major constraint on our ability to share data collected from DOLLY, as dictated by the Twitter Terms of Service, is that we aren't able to share any information about the users or the text of the tweets.

      Delete
    3. Yeah, Twitter TOS is not fun for researchers. :(

      Delete
  19. Would really like to see this without the normalizing based on volume of tweets. What would it look like normalized for population? I bet California wouldn't look like this hate-free paradise.

    ReplyDelete
  20. I am going to venture an answer to the question of why the the authors of the study didn't track Tweets hostile to women/white people/Christians/conservatives/etc.

    They can't do everything. It is impossible to draw a map of all the forms of group hostility in America that get expressed on Twitter, so they have chosen to focus on a few that everybody knows about and generally agrees are bad. That doesn't mean that other groups don't get attacked on Twitter, perhaps unfairly, perhaps in ways that would also make for an interesting map. This study is not the final word on hate. The omission of your preferred group is not itself an attack.

    ReplyDelete
    Replies
    1. To be honest, prejudice against majority groups (white people, Christians, and to some extent Conservatives -- who may be getting called names because of their abhorrent social beliefs rather than inborn traits, which makes it fair game), people who wield significant social influence, so prejudice doesn't really affect them, except it might hurt their feelings, just doesn't really matter so much. It's an inch compared to a mile.

      Delete
    2. Two wrongs don't make a right, Zoe. MLK and Lincoln understood that; it's why the former advocated peaceful protest and the latter advocated the reconciliation of North and South (before his assassination). They both understood that true peace lies not in the transfer of majority power, but in the elimination of power structure altogether (as it relates to racism; I would guess the concept is applicable to other isms but I don't know for sure).

      So just because prejudice doesn't affect the power of majority groups, that doesn't make it OK to ignore the fact that such prejudice exists. Or would you deny that any homosexuals have irrational hatred of heterosexuals? That any racial minorities have irrational hatred of white people? That any women have irrational hatred of men? By focusing on "hate-ism" as a function of "who has national social power", you're perpetuating the very problem that a few extraordinary individuals have tried to eliminate. Most of them were assassinated, so I guess in a gruesome/practical sense I don't blame you.

      In the original "Brian's Song" made in the 1970's, there's a scene where Gale Sayers and Brian Piccolo are told that they will face racism as mixed-race roommates no matter where they travel...from white people in the Southern cities and black people in the Northern cities. Fast-forward to the remake in 2001, and the scene is changed so that the players are only informed of racism from whites in Southern cities.

      Do you see my point? Equality will not be achieved if we ignore one side on the grounds that it "doesn't affect them." Power is NOT the answer; it is the problem.

      Delete
    3. Let me add that I was skeptical of this map when I first saw the link, and I think people have brought up excellent criticisms. But on further reading the map won me over, it's a great concept because it recognizes social media as a (somewhat) measurable indicator of people's prejudices. I'm hopeful to see how they and others refine this in the future, to examine something that would have depended on more questionable methods in the past.

      Delete
    4. "I am going to venture an answer to the question of why the the authors of the study didn't track Tweets hostile to women/white people/Christians/conservatives/etc."

      Did you just lumped women-as-a-group with white people, Christians, conservatives, etc? You lumped women, who have been denied basic citizenship rights for most of this country's existence, with the privileged groups? Are you kidding? Please tell me you're kidding. Please tell me you are not so historically illiterate that you don't know that women have been second or third or no-class citizens throughout this country's history.

      Delete
    5. And you refer to women as "your preferred group." Great. We live in a world where women are raped and then whipped and shamed for it, where girls are shot in the head for wanting to go to school, where women lack basic human rights in countries around the globe and the world looks the other way. Misogny is the single most damaging and violent form of hatred in the world. And to you it's just "your preferred group" and "it might make an interesting twitter map."

      Delete
    6. Njstuckey, you sir or ma'am blew my mind, you put in to words how I've felt for along time. And sophie you need to get a grip with reality. Complaining about how things were in the past and around the rest of the world isn't going to get you very far. I'd be of mind to say the problems not as bad as you think it is. In the company I work for our leadership staff is at least a third women. Now in America and for nearly 100 years we've been legally equal, socially it takes longer but we have come a long way. What exactly are we supposed to do about the rest of the world? Wage a war? Economic sanctions? The best we can do is lead by example, which is what we do. And to zoe, how is being white not an inborn trait, but rather an "abhorrent social belief"? Anyone regardless of gender race or religion can have abhorrent social beliefs. Not just white Christian republicans.

      Delete
  21. Agree w/ W.P. McNeill. I'd suggest that the reason for the relatively small selection of hate-speech types and words comes down to resources. Consider that the creators of the study used human volunteers to pore through 150,000 tweets manually. If they had unlimited resources, they could increase the scope of the study dramatically. As it is, they never claimed that it is comprehensive.

    ReplyDelete
  22. Just saw this floating around the Tumblr-sphere, and have to say I'm so happy to see Humboldt in the URL - my own alma mater! Thank you for doing these kinds of great work (and putting up with the nutters in the comments too.)

    ReplyDelete
  23. This is very interesting and thanks for the great work. It would be interesting if you could do some validation like comparing cancer tweets with SEER data on cancer incidence and see if the county level data matches. Or, perhaps more importantly if you could identify anti-American sentiments and map them to identify areas that may contain concentrations on anti-american groups.

    ReplyDelete
  24. Are many of the same people issuing homophobic and racist tweets? Looks like that might be occurring.
    bob mcconnaughey

    ReplyDelete
  25. This is very interesting, but not very surprising. It validates what I have always thought in regard to racism in the United States.
    I am surprised however, that a criteria for MUSLIM or ARAB was not used in this study. I suspect, that the HATE for this group is at an all time high.

    ReplyDelete
    Replies
    1. P.S. I find your study fascinating and intriguing. Thank you for your work.

      Delete
    2. As with misogynist tweets, I imagine the workload involved in separating out positive uses of "muslim" from negative ones would be beyond the scope of the study.

      Delete
  26. can you tell us more about your methodology, both for discerning which tweets are really negative and for normalization? would you mind sharing your county-level data--not the tweets themselves but the counts for each slur geocoded by county?

    ReplyDelete
    Replies
    1. Yes... and not just the numerator but the denominator (all tweets) too. Free the data, free the code.

      Delete
  27. The "hover" feature (that should show the underlying data at each particular county) does not work for me in either IE or Firefox. Is it broken? Do I need to use a different browser?

    ReplyDelete
  28. I'll point out that the regrettable popularity of that particular slur as unique to Texas is probably a function of the riparian border Texas shares with Mexico (and hence the assumption that most undocumented immigrants in Texas must have waded across), while the other three border states have only surveyed lines in the desert to mark the international boundary.

    ReplyDelete
  29. The methodology behind this map was far from rigorous. First problem is they break the data down by counties however represent each county as a data point, not using the actual boundaries of the county. This causes wildly inaccurate results when you zoom in or out. Second problem is that the map doesn't account for population density or total internet presence, it only counts total number of incidences. 1000 hateful tweets from Harris County, Texas (pop over 4 million) would be represented identically to 1000 hateful tweets from Loving County, Texas (pop less than 100). That is wholly unrepresentative of reality.

    Fix the design problems with the map and you will certainly get results that are more accurate to what is actually being measured.

    ReplyDelete
    Replies
    1. While your first point is well-taken and can lead to some interpretation issues depending on what scale you are viewing the map, your second point isn't exactly correct. While we didn't normalize by population or number of internet users, we did normalize by the total number of tweets in that place, which generally correlates with measures of population. So 1000 hateful tweets from a county with 4 million people would only be seen as equivalent to 1000 hateful tweets from a county with less than 100 people if they had the same number of total tweets, which is doubtful.

      Delete
    2. Mr. Shelton, I'd really like to get some cardinal benchmark information about the underlying data. Perhaps that is inappropriate in a comment line; if you feel so, please shoot me an email at nwilcox@chapman.edu.

      At any rate, I assume that when you say "normalize by" you mean "divide by" so that the quantities coded by map colors are rates that are greater than the mean rate. I would like to know that mean rate, the ratio X/Y, where

      X = all geocoded tweets (summed across all counties) containing slur words that were deemed negative by the student raters (if I understand the various things your team has posted, this is the 'more than 150K' number that is mentioned repeatedly but I am not certain of it and would like to know the exact number); and

      Y = all geocoded tweets (summed across all counties).

      I would call this the "base rate of hateful tweets amongst all geocoded tweets," i.e. the observed likelihood estimate that a randomly selected geocoded tweet in the lower 48, over your time period, is a hateful tweet.

      I would also like to know what the colors on the map actually represent in quantitative terms. For instance, does the brightest red code for ratios at or above some quantile q (greater than the 50th of course) of the distribution of the ratios Xc/Yc across all 3000+ US counties (each indexed by c)? Or do you calculate a std dev from the Xc/Yc distribution and plot (say) +.5sd, 1sd, etc. above the mean X/Y with increasingly warm colors? Or what, exactly?

      Generally speaking, it would be nice if your team would provide a short technical note, say as a pdf, that contains some sample information and procedural information at this level of precision, because some of your readers are going to be quantitative scholars who need this sort of stuff to make fair judgments about the meaning of the map.

      Delete
  30. I have two comments about this.
    The first is that there are MANY fake accounts set up by sock puppets and Google Monkies that could easily skew things to create public sentiment.

    Second, it's easy to tell who hates Arabs, Asians, immigrants, etc. Look at how many friends one person has with different races and nationalities. For example, one white person can have friends who are black, Indian, Jewish, Islamic, Ethiopian, Greek, Asian, Persian, the whole 9 yards and all that jazz, right?

    The other white person has NOBODY else on their account except for the same white people who have each other on their other accounts. With connections from high school, work, church, etc.

    Sometimes there might be 10 minorities in a school of 500. If that minority is popular, everyone would have them on their account.

    I've seen comments and they don't fit a "criteria" so negative assessments do go ignored.

    Nobody needed to do a study for me, I was able to figure it out.

    ReplyDelete
    Replies
    1. "Nobody needed to do a study for me, I was able to figure it out."

      But your results are anecdotal at best, given that you 'figured it out' by examining two caricatures. It's probably best that other people do the studies. Some things you can't just bootstrap yourself into.

      Delete
  31. I would be interesting to see statistics of #tweets/state's population - East coast is much more densely populated and may be skewing the picture (away from the central states or west coast)

    ReplyDelete
  32. is it a percent of negative tweets/total tweets or just # of tweets for region. what was the sampling distribution? thanks! interesting discussion topic

    ReplyDelete
  33. Why was this study published under the title "The Geography of Hate", rather than something more neutral like "The Geography of Various Taboo Pejoratives" or simply "The Geography of Rudeness"?

    I don't see any point in "dumbing down" the useful and important word hate by sloppily attaching it to ALL negative or prejudicial use of language. Speaking as a 41-year-old gay man, I have a suspicion that when a sensational label like HATE SPEECH!! is widely applied to a word as pedestrian as "faggot", it tends to inculcate a Princess-and-the-Pea mentality in young LGBT people, and undermines the lesson in Eleanor Roosevelt's adage "No one can make you feel inferior without your consent."

    I can also attest that, for example, gay Republicans and gay Democrats will sometimes call each other "faggot" in an unmistakably hostile way -- and that in some cases, they really do despise each other's politics, and may even hate each other as individuals. But it's not clear to me that their use of the slur "faggot" is necessarily an expression of anti-homosexual hatred.

    P.S. By the way, I understand that you were using the colors traditionally associated with "heat" and "coolness." Even so, it might have been better to do the mapping with varying shades of orange, green, and purple -- since in recent years, Americans have been accustomed to think that "blue = Democrats, red = Republicans."

    ReplyDelete
    Replies
    1. Hatred (or hate) is a deep and emotional extreme dislike that can be directed against individuals, entities, objects, or ideas. Hatred is often associated with feelings of anger and a disposition towards hostility. Commonly held moral rules, such as the Golden Rule, oppose universal hatred towards another.

      First, I'm pretty sure, Throbert, that if I called you a "fucking nigger" because of your clearly ignorant response to this topic that any person of reasonably average intelligence would both discern the "hostility" and "deep or emotional dislike" I had for either you or what you had to say. I mean, "fucking nigger" isn't exactly an intellectual response or now is it?
      Second, The LGBT community doesn't have a princess and the pea mentality. Experiencing that abuse myself has disabused me of any notion that it's not a very serious issue. I live in the Canadian equivalent of Texas and it's clear that even among "polite Canadians" there are serious issues of systemic homophobia. I never feel inferior when slurred against but I do, rightfully, fear for my safety at times when homophobic hatred is directed at me. Even before I ever began to figure out who I was, due to growing up IN a homophobic home and religious environment, I experienced homophobic hatred directed at me. A friend of mine and I held hands walking down the street skipping in front of our girlfriends as a joke, both being big and burly men and not really understanding what being gay was all about, and more than one vehicle slowed down long enough to yell fag at us in the perhaps 2 minutes we did it for. When the last of the vehicles that called us fag appeared to have 5 equally large men in it and continued to drive away much slower than they had arrived, we thought better of continuing our joke.
      I mean, verbal hatred never turns into physical violence so clearly I am (and was)foolish for being more concerned than you are.

      Delete
  34. I don't really understand why people are attacking each other in comments. Hate is hate no matter who it is directed at. Hatred only breeds more hatred. So yeah, some groups are excluded from this study because of limited scope and resources. Stop complaining that the map doesn't show how hated your favorite group is and focus on something that isn't hate.

    ReplyDelete
  35. I'm sorry, but... http://xkcd.com/1138/

    ReplyDelete
    Replies
    1. I'm sorry, but you might try reading the section of the FAQ we wrote specifically to address this...

      Delete
  36. I read through all of the sections you suggested for those with questions about your methodology and I'm still left with one question that I don't think was covered substantively. I'm a Vermonter, and while I'm more than willing to acknowledge homophobia in my home state, and indeed, have experienced it firsthand, I was surprised to see a high concentration of homophobia around Burlington, the largest city in the state. It seems a much more likely explanation for the high concentration of "homophobic" tweets in Burlington, is the reclamation of the word "queer," as Burlington has a number of self-proclaimed queer activist organizations doing a lot of progressive work. How did your methodology account for positive usage of the word "queer?"

    ReplyDelete
  37. @Smeezer:

    First, I'm pretty sure, Throbert, that if I called you a "fucking nigger" because of your clearly ignorant response to this topic that any person of reasonably average intelligence would both discern the "hostility" and "deep or emotional dislike"

    True enough, but if you instead called me a "stupid motherfucking shithead", that would ALSO express hostility and dislike, yet would not have been flagged on the map, simply because it didn't include a particular keyword!

    Conversely, if someone tweeted "Dan Savage is a disgusting faggot," I assume it would've been counted as "negative" on the map. But could we actually be sure that the tweeter's motivation was in fact hatred for all gay people as a category? Maybe. But it's also possible that the tweeter has a specific animus towards Savage as an individual, not towards LGBT people generally, and chose the word "faggot" primarily because it nowadays has greater shock value than "motherfucker" does.

    In short, the title "Geography of Hate" ascribes a particular type of motivation to the use of these taboo words, yet the students who were "scoring" the tweets were -- at best -- relying on guesswork about what the motivation was.

    Incidentally, I agree that physical violence against gay people is a serious problem and clearly an expression of hate, but I'm not convinced that a proliferation of vulgar anti-gay slurs in Twitter is predictive of such violence.

    ReplyDelete
  38. The overall map of hate looks like a wireless coverage map. Since people tend to tweet from their smartphones and their tablets, this isn't really a surprise. I think what was proven here is that most people will show how hateful they really are when provided a way to instantly broadcast their stream of consciousness without the filters that "polite society" demands they use.

    ReplyDelete
  39. I think that there are two main problems about the maps: first, the choice to employ counties centroids instead of counties areas, second, spots' sizes are proportional do intensity.
    This makes maps to be biased by county dimensions and county density.

    Western counties usually have larger areas than eastern ones, so the number on counties in the west is smaller. If spot sizes were all equal or if data were aggregated by county area bias would be severally reduced.

    In your results different observational scales give rise to different interpretations, and for me that is a serious problem.

    ReplyDelete
  40. The map with "Dyke", I would like to add if you put in common misspellings like Dike, there is a town close to the hottest part of the map in Iowa called Dike, so that one might (I stress MIGHT) be a little hotter for a reason.

    ReplyDelete
  41. So far no one's asked about indicators of Jew hating. Surely that ancient scourge still merits attention, no? What about tweets containing the words, 'k**e', 's***ny','Jewboy', 'yid' or 'hebe'? Those last two are sometimes used among Jews as some blacks use the term 'n****r' so that would need to be discounted. In any case, Jew hating can't have sunk to a level THAT insignificant, even here.

    ReplyDelete
  42. Or, for that matter, use of 'Zionist' as a pejorative. I'll second Dr. King's take on 'anti-Zionism'--with which I believe you are familiar.

    ReplyDelete
  43. There is a reasonable explanation for the use of the term "wetback" in Texas. "Wetback" is a reference to wading, fording, or swimming across the river as opposed to more formal ways of entering another country. The entire border between Texas and Mexico is the Rio Grande River. As a former Texan, referencing the Rio Grande is almost synonymous with referencing the international boundary. In other parts of the Southwest (New Mexico, Arizona, California), it would be linguistically and culturally odd to use the term "wetback" because there is no major river forming the international boundary. Perhaps there isn't an equivalent kind of pejorative term in other geographic regions that don't mentally reference a river as an international boundary.

    ReplyDelete