Skip to main content
Menu
Never Tweet, Mr. President

I’m not embarrassed to admit that I feel a little empty inside if I don’t have a @realDonaldTrump tweetstorm to go with my morning coffee. But I’m decidedly in the minority on this question: Even voters who like President Trump don’t particularly care for him tweeting all the time. A Quinnipiac poll in August, for example, found that 69 percent of Americans — including 54 percent of Republicans — thought that Trump should stop using his personal Twitter account.

But do Trump’s tweets merely annoy voters or can they actually make it hard for him to stay on track and maintain his support with the American public? In last week’s edition of our politics podcast, I wondered if there were any relationship between the frequency with which Trump sends out incendiary tweets — such as his recent rants against the NFL and some of its players — and his approval ratings.

The answer is that there probably is such a relationship — periods when Trump sends out exclamatory or inflammatory tweets have been correlated with future approval-rating declines. But the relationship is somewhat noisy, and the causality isn’t totally clear. So we’d encourage you to read this story as a plausible hypothesis that will need further proof.

Here’s how I came to that (tenuous) conclusion. First, I created a statistic called Trump Weekly Twitter Rage (TWTR), which is calculated as follows:

  • Add one TWTR point for every insult Trump made on Twitter in the past seven days, as according to our friends at The New York Times’s Upshot blog.
  • Also, add one TWTR point for every exclamation point Trump has used in a tweet in the past seven days.

Neither of these is a perfect measure of Trump’s Twitter rage — the definition of an “insult” is somewhat subjective, and Trump sometimes uses exclamation points when he’s announcing boring meetings and not just when he’s angry — so combining them smooths out some of the rough edges.1 I calculated TWTR on days from Jan. 27 — a week after Trump’s inauguration — up through and including this Sunday, Sept. 24:

Trump’s Twitter rage peaked on July 29, shortly after the failure of the GOP’s health care bill, when he achieved a TWTR of 112. (The GOP’s health care bill has failed several times — including once this week — but the July failure was the particularly dramatic one when Sen. John McCain voted against the bill on the Senate floor.) And as of Sunday, his TWTR was 77 and rising. Trump is capable of raising his TWTR by as many as 29 points in a day — a record he achieved on Aug. 7, when he went on a rant about “fake news,” The New York Times and Sen. Richard Blumenthal of Connecticut, among other topics — so he could be a threat to break the record later this week.

The shifts in the president’s Twitter mood seem to be at least somewhat related to changes in his approval rating. April — after a Twitter lull early in the month — was something of a high point for Trump, with his approval rating reaching a comparatively good 42.0 percent by April 20. Early August (after Trump’s late July TWTR peak) was a nadir, by contrast, with his approval rating bottoming out at 36.6 percent on Aug. 7. More recently, a relatively calm period on Twitter in mid-September was associated with an uptick in Trump’s approval rating, although it’s since begun to decline slightly again.

Are these patterns statistically significant? The short answer is not quite, although it’s close. (For the long answer, see the footnotes.)2 In other words, the relationship between Trump’s rage-tweeting and his approval rating is probably negative — more Twitter rage means a less popular Trump — but we need more data to confirm it. We also don’t know about the causality — whether it’s the tweets themselves that could be hurting Trump or the underlying issues they bring up. It’s possible that Trump tweets more at times of political distress, such as after the failure of a health care bill, which could make the correlation somewhat spurious.

So exactly how much the tweets are hurting Trump is open to debate. At the same time, I feel reasonably confident in asserting that the tweets aren’t helping Trump as part of some supposed 13-dimensional chess strategy to rally his base or distract the media. Large majorities of the public dislike his tweeting, and his approval ratings have tended to decline after Twitter outbursts.

But just for fun — don’t take this part too seriously — I decided to run a simulation using our regression model to estimate how Trump’s approval rating would change over time given different levels of Twitter activity. The model assumes that Trump’s approval rating is somewhat mean-reverting over time — it tends to rise slightly after really bad periods and fall slightly after comparatively good ones — although also affected by his tweets.

If Trump never tweeted again,3 his approval rating would gradually rise to about 43 percent, the simulation estimates. Conversely, if he went on a Twitter bender and constantly tweeted at his maximum outrage level — the TWTR of 112 that he reached on July 29 — it would eventually fall to 33 percent. Again, I wouldn’t take any of this too seriously. But Trump’s tweets often dictate news cycles and amplify controversies — and they can even help to spark diplomatic crises and put the president in legal jeopardy. It’s not crazy to think the tweets have had consequences — mostly negative ones — for Trump’s popularity.

Footnotes

  1. There’s no limit on the number of insults or exclamation points per tweet: A tweet with multiple insults and exclamation points can score multiple points!

  2. I built a regression model that attempted to predict Trump’s approval rating seven days later, given his approval rating at the time and his TWTR. For instance, we’d be trying to predict Trump’s approval rating on Aug. 17 based on his approval rating and TWTR on Aug. 10. (It usually takes a week or so for changes in public sentiment to be fully reflected in a president’s approval rating.) The coefficient on TWTR was negative, meaning that insulting and inflammatory tweets tend to predict declines in Trump’s approval rating. But the coefficient wasn’t statistically significant at the 95 percent confidence level using robust ways of calculating the uncertainty in the estimates, meaning that these sorts of patterns could have occurred by chance alone. It was statistically significant at the 90 percent confidence level, however.

  3. Or, at least, if he never insulted people or used exclamation points in his tweets.

Nate Silver is the founder and editor in chief of FiveThirtyEight.

Comments