Cambridge Analytica Used Fashion Tastes to Identify Right-Wing Voters | Hacker News Reader

Cambridge Analytica Used Fashion Tastes to Identify Right-Wing Voters

nytimes.com

161 pointschatmasta125 comments

Unfortunately, stereotypes have utility. No one likes being caught up as a false positive in a stereotype, but when you look at mass market politics / advertising / marketing they cater to charactures of their intended audience because for the most part it seems to work. Every once in a while someone takes it too far and a group gets offended, but let's not pretend that Cambridge is the only one doing this.

Stereotype is probably the correct word, but it's overloaded here.

Stereotyping someone by something that they did not choose is a gateway to prejudice: skin color, eye color, place of birth.

Stereotyping someone by something that they did choose may be very useful for making inference about them, because it is a revealed preference: clothing tastes, food preferences, etc.

Stereotyping someone by something that you assume they chose is also a gateway to prejudice. @NeedMoreTea said it right by mentioning cases where someone might not have chosen their particular attire.

It works on a macro level though. You have to be ok with idea that the stereotype isn’t 100% perfect, it’s just used to get a majority of your target market.

Over the last two years, I have witnessed many demonstrations of normal people totally failing to understand “group x tends to be more y” — “but”, I have seen many retort, “I’m x and not y!”, as if the statement had instead been “all x are y”.

Yeah, you have to be OK with the idea that you are part of system that has systemic discriminatory effects. It still stinks.

It also begs the question of whether we have any more free will in fashion choices than we do skin color.

I can see an argument for that but I don't think it's a valid assertion. People can change opinions. They can't so easily change their skin color.

You can change opinions, but it is unclear whether you have free will over the change.

It's just pattern recognition, we have it because it's good for survival.

Maybe that grizzly bear is nice but I'm not going to try and pet it to find out because the stereotype is that it will rip me to shreds

Looks to me like both are prejudice (even if applied to things usually not called such), and both may be useful.

Why is one prejudice and the other inference?

It was hilarious as an engineer who needed to learn marketing 101 for a startup, its pretty much just stereotyping your customers. "Wendy is a 24-30 yr old professional, slightly insecure, straight, likes wine and dogs."

I covered stereotyping quite a bit in my university studies (I studied marketing) and the broad consensus was that stereotyping is a useful tool because it saves time.

As much as we "should" judge people on the nuance of who they are and their individual character, we are presented with an overwhelming amount of information and we need to make a quick decision.

This is why I've held the opinion for a while now that it's good to dress well. People are going to judge you and put a value on your time based on how you present yourself. Nothing wrong with maximizing your odds of that going well.

True, I use it as a social filter to avoid my time being wasted those so easily deceived

There is something wrong with "dressing for success" -- it makes it easier for frauds to get jobs from talented people. but it's a locally optimal decision, and it's not your fault that people judge your work by your clothes, so it's defensible even though wrong as a broader social context.

I think you're talking about using personas. I'm not surprised that was your reaction, as personas are often misunderstood, even by marketing folks.

The point of using personas is not to stereotype your customers but to establish focus sufficient to permit clear product decisions. In other words the idea is not "all our customers are like this" but "this is the type of customer we care about the most." Or more typically there are multiple personas, which allows you to create a matrix of product features with personas and again, establish priority for which feature to develop when, and how much to invest.

I'm using "product" and "feature" here broadly, so for example even in an ad campaign, each ad can be considered a "product" with "features" like where it appears, what's in the imagery, what the text says, etc.

You can run into trouble with personas if you start believing that they cover the entire range of potential customers. They don't, and again, that's the point: focus. That's why personas are sometimes given names and photos and detailed descriptions, to make it clear they're supposed to be individuals not stereotypes.

Maybe they’re supposed to be individuals, but the problem is that unlike real individuals, you made them up. You can’t learn anything from them that you didn’t put into them yourself. Only a real person with lifelong experience of being themselves can give you insight into what that’s like.

If you are designing for someone, you have to base your design decisions on what you have observed that person doing or saying. By your logic, generalizing from past observations of other people is never going to be enough to make an adequate design.

So either you have to literally prototype each touchpoint (for example) with every possible customer (even the ones who can't articulate what they actually need), or the design is going to have to be based on insight into human nature and (varying) capabilities. Luckily, most designed things are simple enough that generalizing about users in order to discover the "right answer" isn't an unachievable fantasy.

Look, if you take a group of 100 people who are into Warhammer 40k, chances are that at least 80% of them will have played StarCraft too. Collapsing that group into a single person and calling her Wendy helps some people reason about them. If you prefer to think of them a Venn diagrams and that helps you sleep at night, no crime in that.

But you’re not learning that from a persona, you’re learning it from talking to 100 real people. You can put that into a persona if you like, but it won’t give you any new knowledge.

It's very common for a new perspective on knowledge you already have to enable breakthroughs in deriving new knowledge from it. In particular, imagining a person rather than a statistical abstraction will tend to activate useful neural wetware that simulates other humans.

Agree 100%, but I don’t understand how that leads to personas. Just talk about real people instead and you can get all of these benefits.

Maybe if you can find the exact right real person who reflects the statistical reality you've already measured, and if you don't feel it's creepy to obsess over this one person in your target demographic, and you don't end up needing their consent to disseminate information about them. It's much easier and approximately as effective (more effective wouldn't surprise me) to make one up.

That is a bad practice with respect to customer privacy.

A persona is a heuristic model, that’s all. It seems you feel like it’s dehumanizing. Would it make you feel better if we called it “human behavior model 748263”? There is nothing inherently bad about saying “if you like video games, you likely own a gaming PC or a console”? What if you call that association “748263”? What if instead of that number you called it “Wendy”? Does any of this make any difference to you?

This is not necessarily the case and makes light of an entire field. Often times personas are developed by ethnographers, human centered designers, design researchers and are based upon actual users or prospective users. Furthermore, data analytics can be used to generate personas. Personas can be representative of key forks in behavior patterns or beliefs and are a useful tool when making design and product decisions. As written your comment seems to write off all of this based upon your claim that "only a real person" can give you insight. While the reality is that personas are based upon user insights and are often distillations thereof.

Right. It's using demographics to predict product adoption, service sign up, etc. This is finding the best-fit stereotypes.

I’m aware that personas are often based upon insights from real people, but I question the value of repackaging the insights as a new, made up person. It introduces distractions and invites stereotypes. Just use real quotes and observations, statistics if it makes sense, and focus on describing typical or recurring situations or behaviors that are relevant to what you’re using the research for. A hair color or favorite snack isn’t going to be helpful.

What about a disability, though?

"A hair color or favorite snack isn’t going to be helpful."

It's about humanizing the product design process, don't pay any mind to extraneous information like hair color etc. At the end of the day, personas should be representative of the people you are developing products for and that's inherently valuable to the product development process and all the people involved who don't have a chance to be in front of users and get a feel for their needs, which determine how successful your product will be when they are aligned with what you are coding.

"Just use real quotes and observations"

Again, it seems you have a strong baseless bias against user centered design practices. You argue against a strawman of your creation.

> You can’t learn anything from them that you didn’t put into them yourself.

Right. Personas should be backed up by research into real customers and potential customers. "Garbage in = garbage out."

Right. With personas, you're using demographics to predict behavior (and, occasionally, craft your product, which isn't marketing but rather product design).

That is another word for stereotypes.

"Product" is one of the 4-Ps of marketing. With that meaningless jargon out of the way, as someone who attended a non-technical marketing undergrad program, product development was one of the tracks, with a course requirement in the general curriculum and a few related electives.

I didn't say demographics. Personas should be built around shared behaviors, not demographics.

If I'm selling gasoline, I'm going to target my marketing at people who own cars. I wouldn't call that a stereotype.

And who owns cars? It leads pretty quickly to demographics, even if you condition on prior regressands. Older folks who rarely drive are unlikely to be the target of large gas consumption (and thus commodification of personas), hence demographic spoilage will happen regardless.

The fact that humans have attributes does not mean every interaction with them is an act of stereotyping. You’re really stretching now.

So long as the factors are predictive, the stretching is simply narrative creation.

So is the solution here just to hire these personas into your marketing department? Is this what all the marketing interns are for?

It's all ok in marketing until you have "the time" :-) Btw this is what the "marketing gels" at Burberry still refer to the time when chavs got into Burberry check baseball caps.

And this is a direct quote from a mate who works for a big 4 consultancy "gels" being the upper class young women who will be working there.

I suspect that Mr Bannon wearing Barbour isn't exactly taken well by their marketing team.

Gels?

Its a joke about the sort of people who work in high end fashion in London - Sloan Rangers think Princess Diana.

one interesting one was that I took 5000 twitter profile photos and averaged the pixel values by male / female. unsurprisingly, the male average had short hair while the woman long. but what did surprise me was that the shirt area of the males had a blue hue and the women had a red one. its not like anyone in particular was wearing red/blue by sex, but on average they were.

Source?

They are their source:

> I took 5000 twitter profile photos and averaged the pixel values by male / female.

You can likely change the apparent gender by increasing/decreasing contrast with high contrast looking more F and low contrast looking more M

would be fun to see if the high contrast M v.s the low contrast F maintained apparent gender

Cool! Could you share the results?

https://twitter.com/autokad/status/689896734542069760

https://twitter.com/autokad/status/689897031414886400

keep in mind these profile pictures are not centered, some are group photos, etc. its not 'super strong' but it seems evident to me, let me know what you think

It's interesting that the hair-length stereotype was surprising to you, but the clothing color stereotype wasn't. Clothing color is stereotyped even before babies start to grow hair, as a way of signaling gender.

There is a large body of evidence suggesting that psychographic profiling is not effective [1] [2]. So I'm actually not sure that stereotypes do have utility when it comes to marketing. If they do have utility, the 2016 election is certainly not evidence of it. CA itself actually admitted that psychographic profiling - the technique referred to in this article by Mr. Wylie - wasn't even used by the Trump campaign [3]:

"...despite articles still featured on its website touting the role psychographic techniques played in Trump’s campaign, the statement added, “We have always stated on the record that Cambridge Analytica did not have the opportunity to dive deeply into our psychographic offering because we simply did not have the time...."

In short, this Mr. Wylie is hyping up the efficacy of the techniques he developed so he can get paid to speak at conferences like the one he did here and extend his 15 minutes of fame. There would be nothing wrong with his being paid to speak at conferences if his techniques had been used in the event that made him famous, however that is not the case here.

[1] https://www.vox.com/science-and-health/2018/3/23/17152564/ca...

[2] https://www.theverge.com/2018/3/20/17138854/cambridge-analyt...

[3] https://www.buzzfeednews.com/article/kendalltaggart/the-trut...

His technique is the same one that Netflix uses to recommend shows and movies afaik.

Products have a market fit, and those markets are typically segmented by stereotypes. Nothing nefarious here.

Edit: though his technique seems less impressive now. It was definitely effective and stretched ad dollars further. Which is targeting companies value add to begin with. So not sure what the complaint is. Perhaps it didn't alone swing the election it's still a variable. Similar to aviation accidents, it's typically never one big thing it's several small contributing factors. Here too this was one small thing, another is disenfranchised middle/lower class, list goes on.

> It was definitely effective and stretched ad dollars further.

The evidence seems to indicate that not only is the technique not effective, but that it wasn't actually used in the campaign. More than one article, in traditionally liberal publications that would normally be on his side, refers to Mr. Wylie's techniques as "snake oil".

Also remember that we only know about Cambridge Analytica because Mr. Wylie announced to the world that he had created "Steve Bannon’s psychological warfare mindfuck tool". That is a grandiose statement if I've ever heard one. His career since his announcement seems to have consisted of going around and being paid to talk about his claims. So we have one guy, who stands to benefit from such claims, saying all of this in the face of a mountain of evidence from credible news sources that shows that his techniques didn't work and weren't used. It's fair to say that his claims are suspect at best, which makes the idea that CA had any effect on the outcome of the election highly suspect as well.

Everything you wrote is wrong - at least as far as CA's contribution to the Brexit vote. In reality CA used very specific targeted advertising on FB.

https://mobile.twitter.com/Nealb2010/status/1059068463933743...

The point about these campaign structures isn't that they're generically effective. In fact they don't need to be generically effective.

They just need to be able to swing a vote by a few percent to define the outcome.

In the 2017 UK election, the Tory majority - such as it was - only survived because of a couple of marginals returned majorities of tens or hundreds.

The adversarial zero-sum nature of representative democracy amplifies those tiny voter preferences into a monolithic national mandate which does not in fact exist.

This how CA's snake oil works. Where a vote is close and uncertain, it only needs to alter the voting intent of an absolutely tiny percentage of voters. That turns out to be almost trivially easy to do.

I didn't mention Brexit, I was talking about the election that made this Wylie character famous. Nevertheless, again, despite the claims in random Twitter posts here and there, mountains of evidence point to psychographic profiling being largely ineffective regardless of what type of vote it is being used on. “Swinging the vote by a few percent” is a major undertaking. It is not “trivially easy to do,” as you put it, certainly not through scientifically dubious methods that (in the case of the 2016 election) are not even actually used.

Mr. Wylie is a programmer that violated confidentiality agreements with his employer to try to gain notoriety and presumably raise his income. That will make him difficult to employ ever again in any traditional setting, even among those that are sympathetic to his political beliefs. He seems to believe, or at least wants others to believe, that he is today’s Edward Snowden - even though the two couldn’t be more different. I don’t know how well his monetization efforts are working out, but based on the fact that he is now making clickbait-type public statements that can be easily debunked to get media coverage, it seems that it may not have worked out as he planned.

Stereotypes is just another name for abstractions and categorizing, which is how science works all the time.

It's just that there are reality (statistics) based stereotypes and bad stereotypes (e.g. jewish conspiracy and such).

Still, "all people are unique" is BS, more people within certain groups are more alike than different.

X DOES Y because X belongs to group G is always a bad use of stereotyping (except if Y and G are connected by logical tautologies).

But "X is MORE LIKELY to do Y because X belongs to group G" can be absolutely valid, and prove a great tool for marketing, policy decisions, threat identification, and such.

It's also important to keep in mind that the history of "bad" stereotypes usually has somebody intentionally signal-boosting them for their own gain, often financial, like medieval kings cultivating negative public opinion about Jews so they could exile them and take their stuff, but sometimes more indirectly social, like the elites in pre-slavery American colonies portraying blacks as inferior to keep the indentured mistreated lower classes from joining together in rebellions.

Less malicious stereotyping doesn't have that history of being intentionally used to oppress or "other" groups of people in an organized way, and so doesn't have nearly as much offense to it. For example: "white men can't jump", or associating action movie stars or bodybuilders with Austrian accents.

Is the Austrian Bodybuilder a Stereotype, or is it an archetype based on Arnold Schwarzenegger?

That's not quite right. Abstractions are true but simplified and miss some detail. Stereotypes are correlations where the coefficient is treated as equal to 1 for convenience.

Even reality-based stereotypes are bad, because they perpetuate social ills like racist discrimination. Slavery is reality-based economic system. Being true doesn't make it good.

That begs the question, IMHO:

Saying market analysis is nothing new is similar to the argument that police examining phone records is nothing new, therefore mass collection of metadata is not a problem. The changes in technology make the analysis associated with marketing far more powerful - with orders of magnitude more targets, more data, and more analysis applied to each individual - to the point where there is a risk to society of mass privacy violations and mass manipulation.

And when this risk is applied to the political sphere, it becomes much more consequential. As the the Cambridge Analytica whistle-blower Christopher Wylie was quoted in the article, it was used by "Steve Bannon [to] build his insurgency and build the alt-right". Do we want this power, and the people who control it, to control our countries and societies? Probably people buying LL Bean jackets didn't know it would be used in an attempt to manipulate them.

I'm not dismissing the fact that big questions remain: How effective is the manipulation (the biggest question in IT, IMO), and what makes some manipulation dangerous and some just regular persuasion? But these are very serious problems that might undermine things as fundamental as liberty and democracy.

The more PC word is “heuristic”. Not much different than having to bin 240 students into 240 different placements.

A perfect optimization of their preferences would require going through 240! Combinations for each placement round. But that’s a massive number, so you take some shortcuts and get a “good enough” solution for a lot less computing power.

It's not a stereotype. It's a feature in a statistical model. The only reason it receives a lot of weight is that the empirical evidence aligns with it.

That's pretty much a definition of stereotype. Humans too form statistical models in their head, even if they don't usually call them that.

"a widely held but fixed and oversimplified image or idea of a particular type of person or thing"

A statistical model is not an oversimplification.

Of course it is. "Oversimplification" is like literally the definition of "statistical model" in layman's terms. Statistical models are useful precisely because they cut out almost all the details.

oversimplification: "simplification of something to such an extent that a distorted impression is given."

A stereotype needn't have a statistical significance. In fact, it often doesn't.

You're right that it a human equivalent, but most people form their models from statistical samples that are woefully small, biased, and fail to control for other factors... and worse still, stereotypes aren't actually informed by their own observations, but mere hearsay.

Sure, Induction is the basis of knowledge. But, it can’t tell you why something is the way it seems to be. And, it also can’t tell you if it will always be that way. For that you need theories and models that can be repeatedly tested.

For example, if you wake up every day and see the sun rise, it’s safe to assume that it will always rise. But, without understanding the forces that cause the sun to rise (and the forces that cause those forces), you can’t really be sure. And, even then, an unaccounted and unknown force can come along and break the model (see: the black swan problem).

Induction is unavoidable and very important, but it also the root of many logical errors. In social science, there are so many unknown forces at work, theories are very fragile.

it is useful for the self interested.

The game theory but is that if some stereotypr, they will come out ahead in the short term.

If all stereotype then everyone loses due to the stereotype killing off any that buck the stereotype. That is, the stereotype forces its own truth, which may be bad by killing diversity and adaptation.

Obviously allowing some to do it to get ahead in the short term will inevitably lead to everyone doing it, that’s why we should discourage all from doing it.

Stereotypes have some limited utility, because people who believe that stereotypes have value reward you for conforming to them, and punish you for deviating from them. For example, they may only market a product to you if you conform to a stereotype they have of people who would use that product, instead of materially analyzing whether the product would be a good investment for you and how to present the argument, which is harder than just going "you look like one of our customers, but your look isn't complete, let us fill that hole."

Of course, if what you're selling is a look and not useful for anything else, this is a good way to market, but if your product is useful, you can do better.

It's also important to add that CA's techniques didn't work.

Are you telling me that people who buy camo clothes are more likely to be conservative? That's crazy talk. Next thing you'll tell me is people who dye their hair pink or blue are more likely to be liberal.

Next thing you'll tell me is people who dye their hair pink or blue are more likely to be liberal.

I have been curious about this particular recent feature.

Does brightly colored/neon hair indicate the person is part of some particular group?

My exposure to this trend has been primarily with people on the internet, specifically Twitter and it seems to be predominantly women, and predominantly associated with being outspoken on modern civil rights issues.

Outspoken in which direction?

Not liberals of the traditional type. This type of hair is adopted by some females who espouse intersectional views on how society works.

The hair style is both a signal to others and a modest act of rebellion against societal norms. Modest because similar hairstyles have been around for decades in older countercultures.

>modern civil rights issues

That is the most positive way I have seen somebody frame them. I guess they would even call themselves social activists...

> Does brightly colored/neon hair indicate the person is part of some particular group?

Yes. The "I'm non-conformist and quirky!" girl.

Of course, they're "quirky" exactly the same way as all the other "quirky" girls.

This is my social group. Neither I, nor my friends, dye our hair because we want to be non-conformist or quirky — we dye our hair because it looks cute. It’s not at all difficult to think of fashion choices that’d be more distinct, but it’s often the concern that we’ll be seen as “weird” that keeps us from choosing them. I look how I do because I want to be liked and respected, same as anyone making the choice to be fashionable.

I remember a scene from "NYPD Blue" between Sipowicz and his adult son [who wanted to become a cop like his father]. The advice was that the "most basic" thing to remember was "People, Places: The things they do, the times they do them".

If you turn a hard Left at Crazy, that's profiling ... but it's not wrong. Going with the example provided in the scene, it's good to figure out that the night-shift worker at the toy factory is bringing home a toy gun for his kid at 3am Christmas Morning instead of a high-profile over-armed robbery in an otherwise deserted industrial district in the middle of the night during a holiday.

It seems like the "grey line" shrinks and expands at will depending on the context one desires.

This is how marketing works. They could have used music taste or religious affiliation or nearly any of the data add-ons Experian sells.

This isn’t sinister in any way, merely clever as it seems marketers had been behind the times.

There are entire magazines and television shows that have equally strong affiliation. If spend on fashion related stuff was smart, it was because it was a bargain compared to the many other methods.

>"Marketers have been behind the times"

I have to laugh. I learned psychographics from Bill Wells back in the mid 1990s. He had come back into academia in after spending years at DDB Needham creating one of the first psychographic initiatives/programs that had teams of statisticians, social scientists and used mainframes for statistical analysis in support of product design, message refinement and targeted marketing and advertising based on "personas".

One of the fascinating things to me was that Bill told of doing a summer project during his own graduate school experience at Stanford in which it took him a whole summer to manually (using a mechanical calculator) calculate a Principle Components Analysis on some consumer data.

It boggles my mind in some ways that we have a bunch of CS trained start-up guys who are rediscovering both techniques and consumer behavior research that is sometimes 30 or 40 years old and think they have invented something new! OK, I am sounding like an old fart.

Now, I will admit that many marketing executives I have met over the years were not really comfortable with quantitative approaches but there is a whole subset that have been applying data science types of approaches for a very long time.

> Mr. Wylie explained that clothing preferences were a key metric for Cambridge Analytica, whose business was constructing and selling voter profiles drawn from Facebook data.

> “Fashion data was used to build AI models to help Steve Bannon build his insurgency and build the alt-right,” he said.

Cambridge Analytica built the “alt-right”? Even if that term has any meaning (I don’t really think it does), this statement is preposterous. Bannon’s “insurgency”? No spin there.

Not sure why you're getting downvoted. The article is obviously pretty low-quality and technically incompenent; I'd expect better from HN's front page.

Even with the left-wing spin, arguing that CA (or even Bannon) built the alt-right is just laugh-out-loud wrong.

I don’t think we’ll get an answer but the downvotes all came in a span of a couple minutes so maybe some partisan site linked to the HN post or something.

Down voters feel free to explain if you actually have a objection.

> Bannon’s “insurgency”? No spin there.

Bannon himself used the term: "We're leading an insurgency movement against the Republican establishment"

https://www.independent.co.uk/news/world/americas/steve-bann...

I don’t know if this is interesting. It is their job

This sort of thing is going to cause a huge chilling effect in the medium-term as it becomes more widespread. Every nuance of how you act or look is going to get collected by websites and cameras and used to make conclusions about who you are (accurate or not), potentially stored forever.

One can only imagine how China will use these approaches with their planned 570M cameras with facial recognition.

edit: number of cameras was an order of magnitude low, corrected based on: https://www.businessinsider.com/china-facial-recognition-tec...

I wonder what google or reddit know about my political ideas..

I often wonder what HN could infer from my positions on various topics based on my voting (eventhough I don't use it to signal (dis)agreement).

I just ask myself how I, a person who satisfies quite some nerd stereotypes would be classified by these algorithm (in particular for political advertising).

So for example the small number of categories that Google Ads assigns to me (https://google.com/ads/preferences) are correct - but featureless and quite hard for marketers to target.

It's very likely that you just think they're "featureless and quite hard for marketers to target."

For example, this story (from 2012) about Target's ability to predict a woman's pregnancy: https://www.forbes.com/sites/kashmirhill/2012/02/16/how-targ...

Or they got lucky in that case, and we only heard of it because of such luck.

> It's very likely that you just think they're "featureless and quite hard for marketers to target."

At least of the second part, I am pretty convinced. When I see ads, I always think how wrong and clueless the ad network is again of my true interests.

couple years ago i heard that adult content industry was selling pretty precise political profiling of their users. So it isnt just fashion.

Hardly ground-breaking stuff; individuals do this in everyday life as a matter of routine. They even use this tacit knowledge when they attempt to manipulate others: just compare 'old' Chris Wylie with 'new' Chris Wylie.

We use the erc20 contents of ethereum addresses to determine exactly which influencers someone is exposed to and will be exposed to

This is because nobody uses the best practice of avoiding address reuse

These are basically brands

Not really surprising. I catch myself making assumptions about people's politics based on the clothes they're wearing. This data just allows one to confirm.

People who buy Wrangler and LL Bean might vote conservative. Hard hitting stuff right there.

I suppose as per the article we could make a guess about the people that buy green hair dye?

I thought it was interesting. What's the stance you're taking with the anger?

There is no anger. The obviousness is the issue. If anything I wonder if this article is just some NYT’s writer admitting they have no idea what people outside of cities do, wear, eat, think.

Of course we can make guesses about people who buy green hair dye.

Not sure why downvoted, I agree. A 'western' and 'outdoorsy' company tends to have conservative customers, I think any 12 year old could tell you that.

I'd rather see unexpected/unexplained trends, things the average person wouldn't already assume.

I think it actually goes a bit deeper than using certain brands to identify political leanings. That was step one, but they also rated users' openness, conscientiousness, extraversion, agreeableness, neuroticism, depression, anger and vulnerability.[1] Then they showed targeted messages to people they decided were right leaning and vulnerable to their messaging.

[1] https://www.cnn.com/style/article/christopher-wylie-fashion-...

Yea... they did all that... then fell all the way back to “who buys wrangler jeans?”.

Makes perfect sense. I remember seeing something a while ago about an fMRI study which showed that the difference between "Coke drinkers" and "Pepsi" drinkers had more to do with people's sense of identity than with actual physical enjoyment of each drink. We're probably all familiar with the stereotypes of latte-sipping liberals and beer-swilling conservatives. Are liberals physically different in a way that makes them more likely to enjoy latte? Unlikely.

People identify with brands as reflections of their identity and affiliations all the time. This is especially evident with cars. What kind of person drives a Prius? What kind of person drives an F-150? Don't tell me that mentioning those brands didn't make certain images pop into your head. I drive a Mini and love it, but I'd never buy a BMW 3-series (which is built from many of the same components) because of the brand associations. Of course someone who wears Carhartt is going to fit a different political profile than someone who wears Under Armour. It would be foolish not to use that information in targeting an ad or political campaign.

Honestly... who cares??

Agree. This is simply good targeting.

Once again demonstrating that "deanonymized" data is anything but.

So it is a cultural war.

... how? How is an identification methodology a war let alone a culture war?

Because people's fashion choices, food they like, beer they drink.. that's culture. Two opposed cultures fighting over everything? Culture war.

It's another point in evidence of something we can all see.

So what information could they see? Could they see posts and shares? If they installed the app from before? I mean you can find right wingers by looking at all of the right wing pages on Facebook and seeing who is in them. I can't imagine it's that difficult.

Why is this news? Everyone does this, including the nytimes. The nytimes marketing department offers their clients on all kinds of identifying/targeting traits for advertising.

From your fashion to your viewing habits to which news you consuming to even the types of pets you have and everything in between is used. Where you work, where you live, what kind of parenting situation you have, etc.

Isn't that the point of political and corporate advertising? Isn't this why the nytimes and cambridge analytica is in business? Is the obvious "news" now because of trump and the "right wing"?

Considering the DNC spent a lot more money than the RNC the past few elections, there are far more organizations who targeted left-wing voters than right-wing voters. But I guess the nytimes doesn't want to write an article about itself and its ad team and their clients.

As much as I think Trump's idiotic anti-media rhetoric erodes our democracy, when I see the NY Times publishing articles like this, it makes me cringe. First of all, it's a complete non-article -- marketing firms used correlative data to make inferences about political affiliation? No way!

Second of all, instead of being a smart and pointed analysis as to why this might be bad, it somehow draws political conclusions from a fashion conference and displays a large header featuring the green-haired Christopher Wylie making fantastical claims about "Bannon's insurgency" -- even an iota of research would clearly show that today's "alt-right" movement was a decades long project started by Newt Gingrich (among others in Republican leadership) in Georgia, escalated during Clinton's impeachment proceedings, and fully developed during the Tea Party movements that preceded Trump.

But of course, these are complicated political machinations that have many moving parts. Instead, the NYT insults its readers by pandering technically- and politically- incompetent drivel like “Fashion data was used to build AI models to help Steve Bannon build his insurgency and build the alt-right” -- give me a break.

These news don’t make a big splash when other companies use profiling to identify people. This one did because those profiles were used to incite violence. It’s news, just not to your liking.

It's really not news, it's clickbait (because it's controversial). I understand NYT and their like needs to make ends meet, but they need to find a better way.

What violence did it incite?

The whole article reads, to me, as "We can sell you ads on the most tenuous associations" as we're too stuck in our own marketing bubble. :)

Stereotypical difference between a peace activist and Wall St banker, sure. No doubt some more subtle variations too. Comparing mainstream brands? That seems a stretch too far to say the least.

Preferences in clothing and music are the leading indicators of political leaning

Or might indicate nationality and age group far more than politics? Music has long heavily influenced clothing.

opting for Levi’s over Rag & Bone makes a statement about associations and history and opens one up to fashion profiling

Or you were born in the 50s and haven't, yet, heard of Rag & Bone, or you prefer Levi's fit, or they're heavier denim. Or you're British and prefer Rag & Bone, as they started here. Or one was on offer at Amazon.

Or your partner chose whichever they prefer on you.

I don't think it was being used as the sole metric. Rather, I think it'd be used as a weight.

X percent of the population wears A brand, while Y population wears B brand. What correlations can we find in some Z count of reference population? Use that correlation value to infer data where none existed.

Yes, some of that inferred data will be wrong. But a lot won't be. And when you're talking a massive numbers game like millions of voters then you're also talking about accepting that trade-off by hedging against other correlations that can help solidify a potential inference.

I think you’re spot on. I have friend that probably defies nearly every major category he’s part of: age, race, population density, location, home and perhaps even vehicle. But after reading this I realized his clothes do stand out and might be one of the few outward things that don’t suggest “middle aged midwest conservative”. If you need to figure out where to focus canvassing and marketing in a political campaign it’s probably not bad to have a range of variables and metrics to weigh, especially when things are otherwise mostly homogeneous. My friend isn’t going to tell his neighbors he voted for Clinton but you better believe he did.

In those terms it sounds perfectly reasonable. Just as "Preferences in clothing and music are the leading indicators of political leaning" is rather different to "Preferences in clothing and music are two strong indicators, of the dozens of factors we used to infer political leaning". Just like Google use hundreds of factors, with probably a few more significant ones, to decide search placement.

I'm sure inferences can be made in aggregate, but the precision that is implied in the article, from just two indicators is a leap. Shoddy reporting, or a little too much hyperbole from someone giving a conference speech I couldn't say.

It's the sort of thing that gives rise to 30 minutes of comedy in the office when there's been some reason, like a news piece, to view FB, google or other profiles. We inevitably find that half the office is mis-gendered, aged, those with children don't and vice versa, and on. These profiles seem to get most people wrong much of the time, but be worth something in aggregate. :)

Pretty much. I think this is also why self-driving cars are still not mainstream, while we all carry little cameras in our pocket that can literally translate text in real time.

Like... the things we have that work? they work 95% of the time, and for translating menus? that's amazing! and really useful.

Advertising is something where getting it right 95% of the time is way overkill, too. so you can advertise on super tenuous links, and it's fine! totally works.

working 95% of the time just isn't okay for self driving cars, and that is what makes the problem so hard.

>opting for Levi’s over Rag & Bone makes a statement about associations and history and opens one up to fashion profiling

Or all these objections just apply to this or that individual case, but in aggregate the observation is correct.

Chris Wylie didn't work at Cambridge Analytica in 2016. He left in 2014 to create his own company, Eunoia Technologies, which was a direct and bitter competitor. The two regularly competed for clients and CA sued Eunoia, alleging that Chris took both clients and data with him when he left. It was a deeply unfriendly parting.

Therefore, Chris has no insider information about what happened at Cambridge Analytica during or in the run up to the presidential election. Therefore, these claims are creative fiction.

Reply on news.ycombinator.com