New Waves in the Study of the Worst Takes Online

As a Very Serious think tank, we at Data for Progress spend a lot of time thinking about Twitter. While only a small fraction of the US population is on Twitter, nearly every major political figure uses it as a means of interacting with constituents and the media.

But not all tweets are created equal. Some are good. And some are very, very bad. Terrible, horrible, no good, very bad tweets.

One simple, popular way to measure the quality of a tweet is its Ratio -- the number of replies divided by the number of favorites (or, in the ever fluid language of Twitter, likes) it receives. The reasoning behind this is that it takes far more effort to write out a response to a tweet than to simply smash that like button and keep scrolling. While not all replies to a tweet are necessarily negative, the more people take the time out of their day to respond to a tweet, the better an indication it is that a lot of people find it objectionable. The vast majority of tweets end up with far more likes than comments, because Twitter is a virtual Skinner Box that trains us to say things that people like us will like, as opposed to dislike.

The Ratio has provided the Twitterverse with an answer to the question: “What is a bad tweet?” But it doesn’t answer the question “What are the worst tweets?” or at least which the Twitterverse has found most distasteful. Here, we explore new methods to answer that pressing question.

Option 1: The Highest Ratio

Formula: Max(replies/favorites)
Analysis: Well if a higher ratio is bad, then shouldn’t the tweets with the highest ratio of comments to favorites be the worst tweets? This makes sense at first glance, but there’s a problem. When we look at the three worst tweets in the congressional dataset by this measure, we come up with tweets that are relatively low engagement and don’t feel like they deserve to win the worst tweets award. Many invite constituents to interact with the Senator, listen to a radio appearance, or engage with them in some other way -- tweets that are particularly likely to generate relatively few favorites and a smattering of replies that are not necessarily indicative of a bad take.

Outcome:

• High ratio: Yes, by definition

• High engagement: Not at all

• Policy importance: Minimal

• X factor: Not feeling it. Low engagement, boring tweets

It’s pretty obvious that beyond not being the worst tweets, many of these tweets aren’t even that bad. The biggest failing of the ratio is that for low interaction tweets, sometimes inoffensive tweets will get ratio’d more or less by chance. A rigorous ratio analysis will take into account not just the ratio, but the amount of engagement it received.

Option 2: Highest engagement and ratio’d

Formula: Max(replies) for all tweets with a ratio > 1

Analysis: This ranking formula is the complete opposite of the previous one. If the overall engagement is more important than the pure ratio, why not look at the highest engagement tweets with a ratio of at least 1? This certainly identified some bad tweets, but it also flagged some tweets that mostly just carried high engagement, and didn’t have the eye-popping ratios that are characteristic of the truly worst takes.

Outcome:

• High ratio: Not necessarily

• High engagement: Yes, by definition

• Policy importance: Mostly, but some are off topic

• X factor: This is much closer, but doesn’t quite capture the magic of an epic ratioing

Many of these are seriously bad tweets.

But are they the worst? A lot of the tweets here are normal, par for the course Republican hypocrisy and heartlessness. This algorithm seems to answer not the question of ‘What is the worst tweet?’, but rather ‘What is the highest engagement bad tweet’?. A number of tweets that are bad, but not truly terrible make it into this dataset. Does John McCain asking for more followers, or Lindsey Graham talking about golfing at a Trump golf course, truly deserve to make it into the rarefied air of the truly worst takes? Probably not.

We want to find the true standouts of the bad tweets. We need a measure that selects for high engagement tweets that are also high ratio. The tweets that were not only bad, but were monumentally so -- takes that Senators were so thoroughly dragged for that it sent shockwaves throughout cyberspace.

Option 3: The Ratio Richter

Formula: ln(ratio) * log(replies)

Analysis: Taking inspiration from the exponentially weighted Richter scale, which is used to measure earthquakes, the Ratio Richter has quickly become our gold standard for rating bad tweets. This formula multiples the tweet’s the natural log of the ratio (comments / likes) by the common log of the number of comments.

A quick math refresher from Wikipedia: the logarithm is “is the inverse function to exponentiation. That means the logarithm of a given number x is the exponent to which another fixed number, the base x, must be raised, to produce that number x.”

The balancing of the two elements of the Richter -- the ratio and the number of replies is the critical step that makes the Richter work.

Outcome:

• High ratio: Yes

• High engagement: Yes

• Policy importance: Yes

• X factor: Yes

The Ratio Richter does it all. It captures the high engagement, high ratio, policy based tweets that have the X factor we’re looking for. From Bill Cassidy’s shameless lie that his bill wouldn’t take healthcare coverage from people with preexisting conditions to Lisa Murkowski announcing her support for the Paul Ryan tax scam to John Cornyn kind of defending a school shooter(?), these tweets get expose the worst of the worst that twitter has to offer.

What Get’s Richter’d

The next step with the ratio scale is to see what actually gets Richter'd. To do that, we created the Bad Tweet Hall of fame, the list of 100 tweets with the highest values on the Ratio Richter scale.

One thing that the Richter makes clear is that big policy battles really are the things that get the most attention on Twitter. While Tucker Carlson claims that the left is obsessed with Mueller and the Russian investigation, the vast majority of tweets in the Ratio Hall of Fame are about serious policy issues.

In fact, Obamacare Repeal and the Republican tax bill, the two most significant policy battles of Trump’s presidency, make up over half of the tweets in Bad Tweet Hall of Fame.

The next two were the events in the aftermath of the Parkland shootings and the confirmation battles around Betsy Devos. While less explicitly tied to a specific policy, both of these are consistent with an online discourse focused around high profile policy battles.

One concerning thing is the lack of highly ratio’d tweets connected to the judiciary. With Republicans having stolen a Supreme Court seat from Obama in order to appoint a far right ideologue in Neil Gorsuch, we would hope to see sustained pressure towards Republicans on this front. This fits in with ongoing Data for Progress research showing that Democrats significantly underestimate both the importance of the judiciary and the threat it poses to the progressive agenda.

Who Get’s Richter’d

When it comes to getting lots of tweets into the Richter hall of fame, there are three (somewhat overlapping) groups who consistently record the highest readings on the Ratio Richter scale:

• Less extreme Republicans (and especially women)

• Republicans from a blue(ish) states

• Republicans without large online constituencies

It’s not surprising to see Lisa Murkowski and Susan Collins at the top of the list: they are among the Republican Senators most likely to get pressure from the left when it comes to important policy. One might expect two of their male counterparts, Jeff Flake and John McCain, to register similar numbers, but they have 4.6 times less tweets in the Richter hall of fame. Even being a senator doesn’t protect you from the hordes eager to correct a woman online.

The second group consists of Senators like Pat Toomey, John Cornyn and Rob Portman. These are your replacement level boring Republicans who support Trump's agenda publicly so as to attract backlash from liberals, but lack the charisma necessary to offset that animosity with likes from conservative users. The third category is the smallest, containing just Cory Gardner and Dean Heller. These are Republicans from blue states who don’t even pretend to represent the views of their constituents.

Meanwhile, Democrats are almost entirely left out of the Richter hall of fame. The only Democrat with a Tweet in the top 100 is Heidi Heitkamp, who nabbed the 40th highest ratio with her announcement that she would vote to confirm Scott Pruitt. Heitkamp is a moderate and a woman, both of which correspond with attributes that predict high Richter readings among Republican senators.

Importance of the Ratio Richter

A large amount of internet meta-discussion focuses on attempting to discern the topics that a particular political group is most passionate about. Are liberals overly focused on Mueller and the Russia investigations to the detriment of policy issues? Which senate confirmations spurred the most anger online? Are progressives fired up about the judiciary?

While the Ratio Richter doesn’t definitively answer any of these questions, it certainly provides context around the ways these discussions have played out on Twitter. Of the 100 tweets with the highest Ratio Richter, 57% were about Obamacare Repeal or the Republican tax plan. Betsy DeVos’ confirmation was the subject of 7% of these tweets, while Gorsuch and the judiciary were only 2%. Most politically engaged readers would agree that these numbers loosely correspond the relative magnitude of these political events.

Further research needs to be done to determine a link between the urgency with which the voting populace views a certain issue and the Richter of associated Tweets, but it's plausible that the issues with the most highly Richter Ratio’d tweets correlate to voter engagement on an issue, particularly for the most devoted and internet connected demographics.