A Quick Primer on Predata’s US Election Digital Campaign Scores

Aaron Timms

We’ve received a lot of questions over the last few weeks on what exactly Predata’s digital campaign scores for the US election show. Here’s a quick primer to help people understand the scores better.

Let’s start with what the scores are not:

  • polls
  • a standalone probability level of victory for Trump or Clinton on November 8
  • a measurement of “media mentions” for the candidates
  • a measurement of online sentiment for each candidate derived from linguistic analysis.

So, what are they?

The short, technical and probably quite unhelpful answer is that the digital campaign scores we put together for the election are the daily, month-long correlations of Predata’s signals for Clinton and Trump to the overall signal for the US election, which is a composite of all the material in the individual candidate signals plus a selection of neutral/explanatory background material about the election online. The scores, in other words, are signal correlations expressed as percentages.

And what, exactly, are the signals?

The signals for each candidate (Trump or Clinton) are very carefully curated and include only material that directly advances the candidate’s campaign. Most of this material is drawn from official campaign sources (YouTube channels, Twitter accounts) and other sources broadly supportive of the candidate and his/her agenda.

The Clinton signal is, in effect, a pro-Clinton signal, and the Trump signal is a pro-Trump signal. The best way to think of the scores we assign to each candidate is as a measure of how well the candidate’s official messaging is “landing” among the online public; it’s a way to understand which candidate is dominating, from day to day, the online conversation overall about the election.

Of course, some of the online public’s interaction with this official messaging (esp. via YouTube comments, replies to tweets, etc.) will be negative, but the negativity is pretty evenly distributed between the two candidates, meaning the net effect is neutral; and remember that the signals themselves only include material that is positive for the relevant candidate. There’s a full explanation of how we put the signals together here.

So what?

Looking at the headline scores for overall conversation about the election (see chart above), the main point is that even though Trump’s official message has dominated online conversation about the election since early September, Clinton has rebounded sharply since the first debate on September 26 and now “leads” the digital conversation about the election for the first time in over a month. This, combined with the most recent polls and an Electoral College picture which makes Trump’s path to 270 votes extremely complicated, means a Clinton victory on November 8 is now looking increasingly likely.

The whole value of these signals and scores is to offer people a new perspective, with unique data, to understand the fluctuations in this and other elections; Predata’s signals act as a supplement to polling, betting markets, analyst wisdom and other conventional predictive measures of electoral outcomes. Our signals for the US election now point to the same electoral outcome as the polls, but where Predata’s data gets really interesting is where it diverges from the polls, as was the case with Brexit (the polls were wrong, we were right: you can read more here).

We’re living in an era of unreliable polling, as the example of Brexit powerfully illustrated. There’s a psychological blindspot built into polling: people may be embarrassed or otherwise reluctant to divulge their true voting intentions to a pollster. We’re not claiming our election signals are perfectly representative of the US electorate (though it’s a fair assumption that most people looking at and engaging with material about the election online will in fact be American), or perfectly balanced between political affiliation, age, or any other relevant demographic segment. What we do think, however, is that in an era of anti-establishment, outsider politics and unreliable polling, understanding the swings and inflection points in the digital political conversation can offer a useful insight into the “confidential pulse” of public opinion. This isn’t to say polls are useless — more that having direct, unmediated access to the volatility flow of online attention can add a meaningful additional perspective to what we see in the polls. Predata’s election signals offer that access. ⏪

For more information on our US election coverage and analysis, contact Aaron Timms, Predata’s Director of Content, at aaron@predata.com.