An Asymmetric Landscape: The Different Shapes of Blue & Red News
This is the second in a series of posts about bluewhitered.org, a media analysis tool that aims to help put the news in perspective. To start at the beginning, go here.
I am, at heart, an engineer and a data geek. So as soon as I saw all the lovely structured data building up at bluewhitered.org, the first thing I wanted to do was build a bunch of graphs to understand the shape of the news. But in the interests of patience, statistical significance and avoidance of bias, I forced myself to wait until I had a solid three months worth before busting out Tableau.
When I read BWR, the quantitative part of my brain is always asking annoying questions like: Is there a particular color on the bias spectrum putting out an overwhelming amount of content and distorting the dataset? Are the ‘centers of gravity’ on each side more or less symmetrical? Are the blue and red sides pretty much mirror images of each other or vastly different?
This is important because repetition reinforces belief — if you keep hearing a statement, especially from multiple publications, it gains credibility in your mind. A common experience I’ve heard users describe with BWR is feeling that a news story is more real if multiple news Sources are reporting it. So it’s worth looking at just how Blue or Red the overall dataset is, any inherent imbalances that exist, and what shapes may be discerned within. Again — this is 100 days of data — not enough to draw any broad sweeping conclusions, but some patterns are discernible.
Before we begin, a quick note on how BWR works if you are not familiar with it (there’s a more detailed explanation here, if you’re interested). News Stories are categorized as Bluest, Blue, Bluish, White, Reddish, Red and Reddest, based on that Source’s typical historical political biases (as rated by neutral third parties). In broad strokes, Bluest and Blue are Left-leaning, Bluish, White and Reddish are more Centrist, and Red and Reddest are more Right-leaning. The full bias spectrum looks like this:
BWR is primarily interested in news related to US politics, so it searches using the keywords ‘politics’, ‘government, ‘elections’, ‘Trump’ and ‘Biden’. It is also looking for news that is being read by lots of people, so it filters out all Stories with less than 10k Facebook likes, which BWR has defined as a popularity threshold.
Here’s what we are about to dig into:
- How are news Sources distributed along the bias spectrum?
- How many news Stories are published by the various Sources along the bias spectrum?
- How popular are those Stories, and how does popularity vary by bias?
- Are the most popular Stories from the Sources that publish the most?
- How does broader engagement (likes, shares, comments) vary across the bias spectrum?
- What variations do we see by date or day-of-the-week?
So with 100 days of data, translating to ~18,500 news Stories, how many Sources are we looking at, and how do they fit along the bias spectrum?
It’s immediately obvious that of the ~150 Sources that published popular stories during this period, almost half are Bluish. Many of them are international Sources, which often appear to skew slightly Bluish when reporting on US news. Russia Today is a notable exception, firmly in Reddest territory.
How does that picture change when we look at the number of Stories (that received 10k or more Facebook likes) published by each Source?
This is a somewhat different picture — let’s talk it through.
Very little content classified as ‘White’ makes it past the popularity threshold — significantly less as a proportion of total compared to the number of ‘White’ Sources. If you read BWR regularly, this won’t surprise you — the ‘White’ column on several topics is often empty. (This is when BWR is viewed on a large screen, with all 7 columns visible — the collapsed view on a phone, with 3 columns, has Bluish, White and Reddish collapsed into a single White column.) Part of this is, of course, due to the number of sources, but perhaps it also means that more neutral takes on a topic are less popular?
Bluish Stories account for over half of the total — so Bluish Sources are punching above their weight on number of Stories published, and are by some margin, the single largest category of Stories on BWR. Could this be a contributing factor to the concept of “mainstream media” being Left-leaning? The New York Times, classified as a Bluish Source, appears to block the news crawler service that I use, so no NYT data points show up in the graph, and no NYT news shows up on BWR. If they did, it would make the dominance of Bluish even more pronounced, especially in comparison to the Reddish category (with Forbes and The Wall Street Journal), which is less than 1/5th the size of Bluish. Another factor that plays into these things is that sites like The WSJ (and NYT) are behind paywalls, and may consequently have less readers than free sites.
Blue and Red, in the top right corner are similarly sized, dominated by CNN and Fox News respectively, and provide the only point of relative symmetry in the whole spectrum. Blue is ~20% larger than Red, but it’s a closer size comparison than the others.
On the far ends of the spectrum, Reddest is more than 4 times the size of Bluest, and is dominated by Breitbart and The Blaze. Bluest counterparts The Huffington Post and Vox are tiny by comparison.
This is a classic asymmetric bi-modal distribution: Bluish dominates the picture overall, and is significantly larger than Reddish. Reddest dominates Bluest. Red and Blue are similarly sized, and White is the smallest by some margin. It means the overall median is Left of center, and that the Red median is further from the center of distribution than the Blue median. More on this at the end of this post.
Does that translate to the Popularity of those news Stories?
In a word, yes. The shape of the Popularity distribution is very similar to that of Volume, with one exception: While there are ~20% more Blue Stories than Red, there are only ~5% more Blue likes than Red (notice the blocks on the top right are now much closer in size). That means that on average Red Stories got more Facebook likes than Blue Stories.
I’ve highlighted some of the most popular stories in each category — there’s a lot to dig into on the content of the stories here, which I will do in a future post. For now, let’s focus on the overall shapes and the big picture.
Next we put Volume and Popularity together.
For a less noisy picture, I’ve trimmed out the one-offs — so only Sources with 5 or more stories that hit the popularity threshold show up in this graph. This trims some (<10) outliers both above and below on the Y axis, as compared to the group shown.
Fox News and CNN are in a category by themselves — prolific news machines, churning out over 15 stories per day that hit the 10k Facebook likes popularity threshold, with an average of >35k Facebook likes per story.
At 5–15 stories per day are 8 Bluish Sources plus Breitbart, The Blaze and MSNBC. This ‘Bluish Cloud’ accounts for a significant volume of the stories on BWR.
At 3–5 stories per day is a smaller ‘Red/Reddish Cloud’, which includes Forbes and Daily Caller, plus 2 Bluish Sources and The Huffington Post. The rest of the pack is a mix, with Bluish Sources dominating given they are such a large percentage of the total.
It’s worth noting that these Sources all likely produce many more Stories per day — these numbers are just the ones that break the 10k Facebook likes popularity threshold.
Speaking of Popularity, let’s take a look at engagement on Facebook more broadly — likes, shares and comments.
Again, to avoid one-offs I’ve trimmed out outliers with less than 5 stories that hit the popularity threshold here.
So it’s worth noting the Sources that still show as outliers after that. The South Florida Sun-Sentinel (Reddish) spikes on shares and comments, and Alternet (Bluest) spikes on likes and shares.
Some other trends are also visible. On likes, there appears to be a slight downward trend from Left to Right, meaning that Bluest/Blue/Bluish Sources got more likes on average compared to Reddish/Red/Reddest Sources.
On shares, this trend looks pretty much flat, indicating less variation. On comments, that trend is reversed, going slightly upward from Left to Right. That means that Reddish/Red/Reddest Sources on average got more comments on their Stories than Bluest/Blue/Bluish Sources.
This is a relatively small dataset, but could the preponderance of comments on Redder Sources be contributing to the perception of Facebook as a “Right-wing echo chamber”?
Now let’s add Time to the mix.
We see lulls on the weekends, and a mid-week spike in news. Early June (when nationwide protests emerged in response to the killing of George Floyd) was a relatively heavy news month for Bluish Sources in particular. More in-depth analysis on this — specifically what topics are covered the most and least across the spectrum — will come in a later post.
Across the bias categories, news cycle mechanics as related to day of the week appear to be the same.
So what have we learned?
As a reminder, here is the bias spectrum again:
There is a clear asymmetry when looking at the number of Stories published by Source. The overall median is to the left of the middle of the bias spectrum, and when looking at the Blues and Reds separately, the Red median is further from the middle of the bias spectrum than the Blue median. The Bluish category accounts for more than half the content on BWR.
In short:
1. Bluish >> Reddish.
2. Reddest >> Bluest.
3. Blue ≈ Red.
4. White is very, very small.
This asymmetry translates closely to popularity of Stories as measured through Facebook likes, the major difference being that specifically for the Red and Blue categories alone, Red Stories got more likes on average than Blue Stories.
When looking at Facebook engagement more broadly, Bluest/Blue/Bluish Sources got more likes on average than Reddish/Red/Reddest Sources. That trend is reversed for comments, with Reddish/Red/Reddest Sources getting more comments on average than Bluest/Blue/Bluish Sources. Shares appear relatively flat across the bias spectrum.
CNN and Fox are in a category by themselves when it comes to volume of popular content, with a large gap before the next biggest players.
There do not appear to be huge variations or skews by bias category over time, with volumes following a typical weekly news cycle.
We haven’t yet talked about the content in each of these stories — which is its own fascinating story — that will be in a future post. If you’re interested, follow us for updates and you’ll be the first to hear when it goes up.
Until then, keep reading BWR, and making up your own mind about the world around you.
End Notes
- It’s a little odd that Sources like The Guardian and Sky News account for as much of the total as they do. While I only request US-focused news, and those Sources do cover the US, some of what gets in is UK specific.
- Facebook likes as a metric for popularity is tricky, I know. Hard to tell if those are actual readers versus click farms or bots. Also people tend to ‘Like’ articles that provoke an emotional response — so content could skew towards the more provocative. If you’ve got better ideas for how to filter for popular stories, I’d love to hear them.
- BWR uses keywords and Facebook likes as filters to show the news that is likely to be most affected by political bias. Important to note that while non-political news occasionally shows up, BWR does not show ALL the most popular news — if it did, it would be very heavy on entertainment gossip, viral videos and sports.
- There is a detailed explanation of BWR’s methodology here. If you have a question that is not covered, do send it over. Please be nice.
