Trump’s magic word on Twitter

By Mark Whitehouse

Bloomberg View

How can Donald Trump keep his millions of Twitter constituents engaged? A look at realDonaldTrump’s activity suggests a simple formula: Disparage Obama and the mainstream media, and say “America” a lot.

Trump is a far more avid tweeter than his predecessor, adding original posts (as opposed to retweets) at an average rate of almost 40 a week. In the first several months of his presidency, this has created a trove of information that – thanks in part to programmer Brendan Brown, who has collected the missives into the Trump Twitter Archive – can be downloaded and mined for insights into his and his audience’s preferences and priorities.

The president, for example, prides himself on using the best words. So which does he favor on Twitter? Leaving out stuff such as “the” and “but,” the No. 1 word from Jan. 1 through May 10 was “America” (or “American”), appearing in 69 out of 668 original tweets. After that came “big” and “fake,” the latter typically employed with “news” or “media” to describe the New York Times, CNN and other major outlets whose coverage Trump finds unflattering. For example:

And here are the top 10 words by the number of tweets in which they appeared:

1. America

2. big

3. fake

4. media

5. Russia

6. jobs

7. Obamacare

8. election

9. Democrats

10. country

But which words are really the best, in terms of garnering support from Twitter users? To get a sense, I did a kind of statistical analysis – called a regression – to see which of the 16 most-used words (or groups of similar words, such as good, better and wonderful) were associated with the most combined retweets and favorites, responses that tend to indicate approval.

One caveat: The use of specific words is just one of many factors – such as time of day and capitalization – that can affect the popularity of a tweet, and probably not the largest. Taken together, the 16 words and word groups explained only about 7 percent of the variation in combined retweets and favorite counts. And most of the individual words and word groups didn’t have any clear-cut impact one way or the other.

That said, some did stand out. The biggest was “Obama,” which adds a very roughly estimated 41,000 to the retweet-and-favorite count – a pretty big deal, given an average count of about 108,000. Naturally, Trump almost always used the name in a negative context.

“America, American” and “fake” were also associated with higher counts. Three word groups – “ObamaCare, healthcare,” “Foxandfriends, Foxnews, Fox” and “Republican, Republicans” – were all associated with lower counts.

All told, the picture suggests a Twitter constituency that isn’t terribly interested in policy and party politics, but is highly patriotic and loves to see Trump take jabs at Barack Obama and an elitist media.

So what’s a president to do with this information? Well, if – as my Bloomberg View colleague Cathy O’Neil has posited – Trump actually operates like a machine-learning algorithm designed to boost attention, one would expect him to provide more of what has worked in the recent past. There’s no clear trend on that so far, but it might be worth revisiting.

Distributed by Tribune Content Agency, LLC.