There are some really nifty graphs here but the conclusions are highly problematic and misleading.
Oliver Roeder, Dhrumil Mehta and Gus Wezerek for FiveThirtyEight:
If the replies stack up, outpacing the retweets and the likes, you may have a problem. Your tweet may be a bad tweet.
“The lengthier the conversation” sparked by a post, “the surer it is that someone royally messed up,” Luke O’Neil wrote recently in Esquire. “It’s a phenomenon known as The Ratio.”
The idea of the ratio of liked tweets, retweets and comments as a measure of whether a tweet is good or bad is overly simplistic. In fact, the idea of a tweet either being good or bad without understanding its intended outcome and measuring it against that outcome is plain silly.
They're many reasons why a tweet might get replied to with greater frequency than liked or retweeted.
One reason might be that the tweet was controversial and the intention was to create controversy. Another might be to get attention, or maybe distract attention away from something else. Perhaps a combination of these things. And any of these things may be a goal of the tweet and therefore — regardless of The Ratio — it wouldn't be bad in context of its intended outcome.
So what can we gather form this analysis? It certainly seems to tell us is that Trump is highly controversial — but we all knew that didn't we? Also, that Republicans more controversial than Democrats.
Or maybe it tells us that there's a high degree of discontent and those who disagree do so to such a great extent that they want to voice their opinions. Remember there's no dislike button on Twitter so when one dislikes something all they can do is comment.
Like the original focus groups were intended to determine what to research — not be the research — maybe this analysis can be used to point toward what we should be looking at.
Analyzing the ratio of positive to negative comments might yield some more insightful results but still the problem exists that commenters aren't necessarily representative samples of a general population. And AI driven sentiment analysis has great difficulty understanding sarcasm which would certainly be prevalent in these comments.
And even if this were valid and reliable analysis and could identify who the "worst tweeter's in politics are — who cares? Do we have more important thing to concentrate on.
I honestly have a hard time believing that FiveThirtyEight isn't aware of the problems I noted above. Seems to me that the reason for publishing this article is simply to sell more advertising even if the article itself misleads the public. This is a shame.