Giving our data to advertisers is the price we pay for ‘free’ internet services. This includes Facebook, Google Search, and more. Every day, our actions on these platforms get gathered up and made available to advertisers. Our profiles are full of hidden data. The data gets used by politicians, corporations, and even small businesses. This data can’t identify you personally. However, it is visible in aggregate to anyone willing to open an ad account.
What many people don’t realize is just how targeted this data is. It can target just Bay Ridge! Today we decided to turn the tables and use the data for good! We used Facebook’s ad data to conduct an analysis of Bay Ridge’s social media behavior!
How We Found The Data
So how did we get these numbers? We created numerous ‘audiences’ in Facebook’s ad manager. Each audience targets an area including all of Bay Ridge (and Fort Hamilton). Then we refined these audiences with what Facebook calls ‘interests’. These interests are auto-generated based on your behavior. An algorithm tracks your likes, follows, views, comments and more. It then assigns you into a ton of different categories. We can then see how many local residents fall into those categories!
How Facebook determines “Interests”
If you have a Facebook account, you can actually see what categories you fall into by checking this page! You might be a bit surprised at what the algorithm thinks you’re interested in. That’s because unlike some data, like age or education, your interests aren’t self-reported. They can be generated by an accidental click on an ad, lingering too long over a video, or liking a friend’s post containing an errant keyword.
Don’t take our word for it. Check out your own profile! This is what feeds into the ad system… and it’s probably going to look weird. That’s because the algorithm casts a wide net. Interests get generated by accidental clicks on ads. They can trigger by lingering over a video. It can happen when you comment on post containing a certain keyword.
That’s because the data doesn’t need to be accurate… it just needs to be accurate enough for advertisers to want to use it. And that means the algorithm casts a wide net.
So because this data is chaotic and messy…
Data Science Warning!
Take this analysis with a ton of salt. This data wasn’t intended for sociopolitical analysis. Facebook has every incentive to over-inflate it’s data. The bigger each category is, the more advertisers will want to use them. The numbers also certainly include fake accounts, bots, and more.
So why are we doing this analysis? Because it reveals what Facebook thinks Bay Ridge is like. In some instances (like news sources) can provide a unique window into how information moves through Bay Ridge’s social media spaces. Without further ado, here’s what we found.
How many Bay Ridge residents are on Facebook?
Yep, there are as many people on Facebook as you’d think. About 77% of all Bay Ridge residents aged 18+ have accounts. There are a few things to keep in mind though. First, we can’t tell how often people log in. Second, location tracking is weird. You can disable it on your phone, so those people might not be in this study. If you use Facebook from a desktop though, they are probably getting your location anyway. That’s reported to Facebook by your service provider, like Verizon.
Even accounting for these flaws, one thing is clear. A good portion of the neighborhood is online and using Facebook.
Data Geek Stuff: We’re assuming that the overall population of Bay Ridge is about 80,000. We got that from the 2010 Census, which under-counted us… we might have a population as high as 90,000 or more.
Our first surprise: local Facebook hasn’t been taken over by your grandparents. Think-pieces about older adults getting addicted to social media are overblown. In fact, Millenials are by far the plurality of users. They are larger than any other two generations combined.
However, Gen Z is growing. The first Gen-Z members graduating from college! Within another year it’s very likely that Boomers will be in the minority… at least on Facebook.
Data Geek Stuff: We defined our generations in the following way: Gen Z, 1995-2015; Millenial, 1980-1994; Gen-X, 1965-1979; Boomer, 1944-1964. Also, you may notice that the total for this data set is slightly larger than the total Facebook user number from the previous graph. That’s because we included ages 13+ instead of 18+ in order to not undercount Gen-Z. This is the only part of the analysis where we include the 13-18 age range.
Local Facebook users are pretty educated
Another hopeful number! On Bay Ridge social media, lots of users are college graduates… nearly half of those who self-reported their education level. Check our Data Geek Stuff caveat below, but it syncs up with census data.
Data Geek Stuff: There were 32,000 people in our Bay Ridge audience sample who didn’t self-report their school in Facebook. The graph above represents just below half of the 62,000 Bay Ridge Facebook users we had in our ad audience. It’s more likely for users with higher education attainment levels to self-report their schools, but we still see some good response rates, especially among high-school dropouts. It’s still probably a reasonably representative sample.
Facebook’s ad categories change fast. Last year, they allowed ads to target political leaning. This has been replaced by “likelihood to engage with political content”. Ad categories are always in flux.
One new category targets users based on where they moved from. This allows Facebook to target expatriate communities. The data, however, is self-reported. It’s based on what cities you list as hometowns in your profile.
There are a few interesting takeaways from the data:
- South and Central-American immigration into Bay Ridge is high. Expect a boom in the Hispanic population in the 2020 census.
- Immigration from Arab countries is… missing. Facebook is one of the most popular social networks in the Middle East. So it can’t be a lack of users. One possible explanation: users are not self-reporting their home towns. If this is due to fear of surveillance, it’s a big problem. People might not self-report it on the Census!
Data Geek Stuff: This map’s numbers, like the rest of this analysis, get rounded up to the nearest 100. This is something that Facebook does when representing the possible audience size. As such, take values that come in at exactly 100 with a grain of salt: they might represent ninety people… or five.
Political engagement on Facebook
Most of our readers are progressive activists. So this might be a shock… Facebook users in Bay Ridge don’t engage with political content. All those flame wars? It’s likely to be the same few people. Those endless comment threads? They get seen by a small fraction of Bay Ridge residents.
Less than a quarter of users are “likely to engage with political content.” That doesn’t mean that users aren’t interested in politics. It just means they are less likely to engage… that means joining political groups, clicking political ads, or engaging in political arguments online. It even means they aren’t likely to click open a comment thread.
What if we break this down by political viewpoint? Turns out that liberal-leaning content gets much more engagement (both positive and negative)!
This means a few things when it comes to social media strategy:
- Conservatives get larger audiences by co-opting liberal posts. This means liberals have a bigger incentive to block out toxic comments. Curating your feeds keeps people from leeching views. It’s also good for your mental health.
- Conservative content gets ignored. Low engagement posts leave people’s feeds faster. Facebook wants to keep people using the site… that means they show you exciting stuff first.
- Moderate content includes posts from reputable news sources. They generate double the engagement of alt-right content! Trump-style hyperbole isn’t as effective in Bay Ridge as it is nationally. At the very least, it doesn’t get likes and shares.
The Trolls are feeding you
This data helps explain politicians like Councilman Brannan. They have a lot of influence on Facebook despite toxic comment threads. It turns out nobody is opening the comments! All they’re seeing is the original post as they scroll through their feed. And those toxic comments actually end up making the post more viral!
The data also helps explain the “de-platforming” conspiracy theory. Conservative users might feel they are getting censored because their posts get fewer views. But that’s because when people DO view conservative posts, they don’t engage. That creates a feedback loop which pushes their content down. Facebook isn’t censoring right-wing content. It’s just that progressive content draws more attention.
Here is a better metric to understand political support in Bay Ridge. Don’t track who is likely to engage with political content… track who gets flagged as “interested” in various political parties.
If you check out the graph, it seems more accurate. It isn’t even close to perfect though. Like we mentioned before, “interested” has lots of false positives. Still, it roughly corresponds to actual party enrollment. It also gets closer to representing how people voted in the last election.
Data Geek Stuff: Socialism and Democratic Socialism were not targetable values, nor were any synonyms. We suspect this is intentional on the part of Facebook. They’re silently eliminating controversial targetable audiences in the lead-up to 2020. As a side note, the Conservative Party was too small to register as a targetable category.
A peek at the 2020 Election
We aren’t limited to political parties. We can check out people too, especially national figures. The graph below shows US Presidents and how many interested users they have! You’ll see that Obama is staying in the lead, though Trump isn’t far behind.
One important lesson is in George Bush Sr.’s numbers. They’re low… but not because he’s dead (look at Reagan). Facebook simply can’t track interest if the subject doesn’t have a social media presence! Reagan is still invoked in comments and content. Bush Senior? Not part of the social media lexicon.
Breaking down hate-follows
Another important point is that “interested” doesn’t mean “support”. In fact, in the graph above, we suspect that Trump’s numbers are over-inflated. He’s dominated every news cycle for four years. It’d be hard to NOT be flagged “interested in Donald Trump” at this point.
Here’s an example of how hate-following can make Facebook mark you as “interested”. We added a new data layer below. It shows how many Obama, Bush and Clinton supporters also get flagged as interested in Trump.
You might think the hate-follow boost might work both ways. You’d be right… but not nearly as much.
Half of all Obama supporters were “interested” in Trump. But only 1/3rd of Trump supporters were “interested” in Obama. This implies that progressives hate-follow conservatives more than conservatives hate-follow progressives.
It also implies Trump supporters are in a bubble. A lack of two-way false-positives means they aren’t being exposed to opposing political viewpoints.
It also means that Trump’s 11,000 “interested” users is exaggerated. Roughly filtering out hate-follows… it is probably closer to 9,000 supporters in Bay Ridge.
The 2020 Primary candidates
This theory gets even more support when you see the 2020 Democratic candidates. They have “interested” users who also seem to be hate-following Trump.
You’re probably more interested in the overall numbers. Sanders is in the lead, with Biden trailing, and Warren lagging behind. We were expecting to see a much worse showing by Biden, who doesn’t have the social media presence Sanders does. That’s because there’s another factor in play:
We can’t tell when someone became flagged as “interested”.
Our interest profile accumulates. It requires user intervention to purge. That means Sanders’s numbers represent both his campaigns combined. He did win Bay Ridge in the 2016 primary, after all.
By the same token, Biden gets a boost from Obama’s presidency. Lots of people enjoyed those “Uncle Joe” memes. Thus, Elizabeth Warren’s lower numbers start looking more competitive. Her two other major opponents were pre-populated into the ad databases.
Let’s look at the bottom of the list: De Blasio and Bloomberg. They are doing far worse than their numbers indicate, for similar reasons. They have home-field advantage for New York… but no boost in interest. Want to know where Kirsten Gillibrand is? We took her off the graph because she had six times the support of anyone else! It skewed the results so much you couldn’t see the graph properly! De Blasio and Bloomberg should have done that as well. They didn’t.
Data Geek Stuff: Delaney, Bullock, Sestak, Steyer, and Hickenlooper did not have enough overall interest nationwide to be a targetable category. De Blasio and most other candidates didn’t have a large enough sample size to estimate how many of their users were also interested in Trump.
How Political are otherwise non-political groups?
At this point we were having too much fun. So we looked at how political random affinity groups were. We looked at groups usually portrayed as voting blocs… and groups that were very much not that. The results are interesting.
When you compare traditional voting blocs to other groups, something weird happens. Non-political groups (like Role Playing Gamers) have similar proportions of politically “interested” users. Traditionally political voting blocs, like Catholics or environmentalists, have 50% of their users flagged as interested in politics. Dr. Who fans, if anything, are more political.
The implications are intriguing. Earning the trust of politically-savvy Heavy Metal fans might have a bigger electoral impact than earning support from law and order voters. Electorally-engaged video gamers are a bigger group than politically-active car aficionados. And cyclists have a far larger representation than census data would suggest… the census only counts cycling commuters. This data counts anyone who has liked or commented on cycling groups, bike shop ads, etc.
This means a new kind of political advertising is possible: culture-targeting. Instead of aiming at an issue, you aim at relatability. It’s probably already happening to help spread memes.
Data Geek Stuff: You might wonder why we left off issues such as student loans forgiveness and affordable housing. Facebook is actually cracking down on advertisers targeting these groups, which is a good thing. As a result, we can’t use these categories in our analysis. We did attempt to include anti-abortion advocates, but the sample size was too small at only roughly 200 interested users.
Also, for clarity, the groupings above are not exclusive. There are only 17,000 local Facebook users that are categorized as intersted in politics out of 62,000 total. There is clearly alot of overlap in the graph above. Video Gamers can also like pets. This indicates that there aren’t as many single-issue voters as you might think: we’re complex in our interests. It also means there might be lots of unused or fake Facebook accounts out there… seriously, who doesn’t like pets?
Fake news exposure
Facebook is very tight-lipped about how much fake news gets onto our feed. But what about regular news? Reading an article, liking a headline, or sharing an article is easy to track. In fact, it’s so easy to track we were able to break down local interest in fifty-nine news outlets across the political spectrum.
This list gives a good picture of what news sources people are clicking on, and the results are encouraging.
Where we get our news…
The most popular news sources are generally high quality. Yes, Fox News is #2 on the list, but it is tied with the NY Times. Fox is also followed closely by more reputable conservative sources like the Wall Street Journal. Junk news and clickbait sites like The Daily Caller and TheBlaze fail to crack 1000 users, and get beaten by left-leaning sources like HuffPo and BuzzFeed.
Finally, the overall number of local users interested in the news is quite high. CNN, Fox, and the NY Times each have the attention of at least 1/6th of every Bay Ridge Facebook user.
Data Geek Stuff: Wherever possible we defined news source bias by using the Media Bias Chart from AllSides. We assigned a few local outlets which weren’t in the list, like Gothamist and Time Out New York, to the “Left” group. You might also wonder where Brietbart and InfoWars are in this list. The answer is: we have no idea. Facebook has eliminated them as targetable categories for ads, which is very good thing.
What talking heads do we listen to?
If we can track news outlet engagement, we can also track which pundits generate the largest interest. We expect fewer hate-follows here since this relies on social media engagement with the pundit’s own pages and fan groups.
You can see that right-leaning pundits have much more of a cult of personality than left-wing pundits, with overall larger followings. However, it can be firmly said that Bill O’Reilly, at least locally, is no longer a thing.
We have zero data available on Joey Salads.
There are more admins than you’d think
We are holding off on analyzing local Facebook groups for future analysis because we can’t target groups we aren’t admins of. However, we can see how many people are admins themselves. And Bay Ridge has a ton of page and group moderators.
This implies that many local users are using Facebook to help organize small, private groups or to advertise a business! Facebook may not be as casually social as you’d think: for many users, it’s part of a job.
Bay Ridge is very unlikely to come to events
It’s common knowledge that when a Facebook user says they are “going” to an event, they mean “maybe”… and “maybe” means “no”. But the data reveals something even worse than that: most people don’t use Facebook events at all. Very few of our neighbors get flagged as being likely to respond to upcoming events.
Maybe spend your time elsewhere if you want to drum up support for your next fundraiser or rally.
That’s it for now! We really had a blast digging through all these numbers. Leave us a comment, email us, or contact us on social media if you have any numbers or metrics you’d like to see in a follow-up article!
Data Geek Stuff: If you’d like to recreate our study demographic, here’s the info you can enter into your Facebook Ad Audience:
Dates Studied: 12/1/2019, 12/2/2019
Who: People Who Live In This Location
Include: 40.6278,-74.0306 +1mi, 11209, 11252
Exclude: 40.6199,-74.0008 +1mi, 40.6254,-73.9985 +1mi