Much has been made about how US President-elect Donald Trump won the Blue Wall – states that typically vote Democrat – with victories along the Rust Belt – states that do a lot of manufacturing.

Meanwhile, much more has been made about how traditional polling failed to predict the outcome and some ink was spilled on how social media data proved to be a better indicator.

To understand how Trump pulled the rabbits out of his hat in the key swing states, we took a deeper look at Twitter engagement data for Michigan, Wisconsin and Pennsylvania during the 2 weeks preceding the election.

120116_bluewallblog_draft5

As you can see, in all 3 states ­– which had voted for the Democratic Presidential nominee in each of the last 6 elections before going Republican in 2016 – Clinton held a lead the week of the October 24th but acquiesced to Trump “big league” over the final 7 days of the race. Perhaps if Clinton had been looking at this same data, she’d have pushed harder in these critical states over the final days of the campaign.

For his part, Trump seemed to key on this insight as reported by Forbes: “When the campaign registered the fact that momentum in Michigan and Pennsylvania was turning Trump’s way, Kushner [Trump’s son-in-law] unleashed tailored TV ads, last-minute rallies and thousands of volunteers to knock on doors and make phone calls.”

Now it’s worth noting that our sample here is relatively small (anywhere from 5k to 70k tweets per state per week) when you consider all the engagement happening across Twitter. The reason is we only included tweets that were geo-tagged – ie, the person shared their location when posting – and that is not a common behavior on Twitter.

However, this is still much a bigger sample size than most political polls which net out between 1,000 – 2,000 respondents. More importantly, social media data is more accurate as it reflects actual observed behavior rather than self-reporting. In the case of this year’s highly-controversial candidates, many people were reportedly embarrassed to admit – especially speaking live with another human being – who they were planning to vote for. Furthermore, we strip out passive signals such as follows and focus on active engagement like mentions and retweets.

As we saw with Brexit earlier this year and the Egyptian Revolution back in 2011, social media – and Twitter in particular ­– can be a great tool for understanding the sentiment of the populace. When it comes to the 2016 US Presidential Election, our analysis of Twitter engagement shows how Trump slipped through the Rust Belt and climbed over the Blue Wall on the road to the White House.