What Presidential campaign videos are Americans watching on YouTube? A Data Visualization from the Google News Lab.

Google Search data gives us a unique lens into what people are most curious about. Throughout the 2016 elections, we’ve looked at Google Search data to reveal the top searched questions on the candidates, spikes in search interest in voter registration, and more.

What people watch is also a way to understand how voters are perceiving the campaign’s biggest moments, whether it’s President Obama endorsing Clinton or Trump announcing Pence as his running mate. That’s why we wanted to look at how users were following the 2016 Presidential elections on YouTube.

We took a look at the most highly viewed Clinton and Trump videos to tease out local and national patterns. Alongside Matt Daniels, his team at Polygraph, and with consultation from Alberto Cairo, we built an interactive that shows which Presidential candidate’s videos are being watched on YouTube across each county since 2015.

We developed an interactive comparing views for the Presidential candidates on YouTube in each US county.

Exploring the interactive reveals a number of fascinating insights. For instance, we discovered that for much of the primary season, Trump’s most popular video was this one about his Boeing 757. And as early as October — four months before the first primary — Clinton’s views were mostly in Iowa and New Hampshire.

In October, Clinton’s views were mostly in Iowa and New Hampshire.

This pattern is something that stood out for Daniels. He pointed out that it suggests that candidates are “driving viewers to their videos, either via promoting on their social media or paid advertising” and that this was “especially clear in swing states.”

This was our first interactive exploring YouTube data by location and time — and we learned a few things along the way:

1. Let users guide their journeys through the data. We wanted to make this interactive easy for users to explore and empower them to discover interesting patterns. Data can be more impactful when a user uncovers a trend on their own. By letting viewers control the timeline and the zoom into the interactive, we enabled viewers to control their own experience.
2. Get and incorporate continuous feedback through the development process. We went through several versions of the map, but along the way realized that we needed some fresh eyes on our work. By sharing this map with our partners at Washington Post, we were able to get their feedback and find what was most interesting to them. For instance, it wasn’t until the Washington Post had played with the visualization that we realized that adding in major cities on the map would help readers get a sense of the geography.
3. Give data a chance to breathe. When we first built the interactive, we developed it to see how the elections played out on YouTube leading up to the conventions. Post-conventions, we decided to refresh the interactive every week here. We soon realized that it takes time for some stories to emerge. By giving our data time to breathe, other stories began to unfold. In September, Trump began dominating in YouTube views in key swing states. And later in the month, Clinton received her highest number of YouTube views in all of 2016 after releasing this video. These were data points that surfaced because we kept our interactive live with refreshed data.

In September, Trump became the most-viewed candidate in key swing states.

There are also a couple of things we’d do differently for next time:

1. It’s important to address the concerns of mobile audiences. When we designed this visualization, we wanted an interactive with data by county, by candidate, and by day. Combining over 3000 counties, 2 candidates, and almost 300 days meant nearly 2 million data points. That’s a lot of data! This meant longer load times, especially for mobile. This lesson reminded us to think mobile-first when we come together to prepare our designs. Could we have focused on weekly data instead of daily? Or key moments in the election instead of the last year? These questions were important ones to weigh as we balanced out the tradeoff between more data vs. faster load times.
2. Make the data easy to read. For this kind of interactive, one question that often comes to mind is how a viewer’s county is performing. We could have created the interactive to make is easier for a user to zoom into their own county, similar to this one from the Upshot, and help viewers immediately read and connect to the data.

With 20 days left until Election Day, we have over 100,000 data points left to add to our interactive. Follow along with our data visualization and discover what other stories will surface as we get closer to Election Day.