There's growing scrutiny of social media buzz metrics related to the 2016 presidential campaigns.
Analysts, the media and campaigns want to know if social media buzz has any predictive value, both in aggregate and as it relates to the voting propensities of individuals. But when it comes to assessing this data, the decision regarding which elements to look at is pivotal to finding insights.
The most glaring issue related to the consideration of social media buzz is that the campaigns, the media and the analysts (for the most part) ignore the vast majority of what's happening on social media.
The most important categories for breaking down social media buzz data are:
- Point of Origin - Social media data measures the content and impact of someone's initial post on Twitter, Facebook or other social media channel. With respect to campaign-related social media data, the most important split is to look at posts that
- Originate from campaign-owned channels - These include official campaign channels, the candidate's personal channels and the channels of any staff or surrogates speaking on behalf of the campaign.
- Originate from an Independent Expenditure channel - These include posts from SuperPACs, 501C(4) non-profits and other political advocacy organizations endorsing and promoting a specific candidate.
- Originate from the general public - These are grassroots, most likely organic mentions of candidates by individual people, whether they are citizens or likely voters, or not.
- Sentiment of post - With respect to the candidate, is the social media post
While sentiment is ultimately a crucial metric for assessing the implications of social media buzz for candidates, measuring it is incredibly difficult. Given the sheer volume of posts about the candidates, assessment of sentiment would need to be done by a computer. And while computers are incredibly powerful tools for counting words and phrases across the hundreds of millions of posts about the candidates, they have severe limitations assigning meaning to those posts, especially sentiment.
For example, computers are very bad at identifying sarcasm - and we all know how much sarcasm is being used to talk about the candidates. Computers are also challenged when it comes to identifying local colloquialisms. For example, while the word "wicked" is generally considered a negative word, but in New England it's typically used to express a positive sentiment.
The challenge of sentiment analysis, while an essential challenge to overcome, is not the focus of this article. Today's focus is on the origination of social media posts and why it matters.
I recently attended a panel about Voter Persuasion Modeling hosted by George Washington University's Graduate School of Political Management at the Microsoft Innovation Center in Washington, DC. The panel included the voter data modelers for Ted Cruz, Marco Rubio campaigns, as well as the analyst from the Democratic National Committee (DNC), among others. The panelists talked about how they were using incredibly rich datasets that included survey data, web cookie data and contextual data to test nearly 200 voter behavior models, often nightly. The results of these models were used to guide voter outreach efforts in order to maximize the vote for each candidate.
What was strikingly missing from these models was data drawn from what people were posting on social media. And given the daily fluctuations they were studying, that omission seemed to be a lost opportunity, at best - and a serious problem, at worst.
When I pressed the panel as to why they weren't using social media data in their models, the answer was somewhat unclear. But one thing was clear, when pressed to think about the question, the answers they provided focused on data related to social media posts originating from the campaigns.
"Yes," they said, the campaigns paid close attention to what was getting traction on their Facebook, Twitter and other social media channels. The consultant from the Cruz campaign further pointed out that their campaign had spent more on social media promotion than any other campaign.
This pattern of analysis is common in the news media, too. And while there are some very good analyses of how the campaigns own posts are doing on social media, there is much less analysis about what other people are saying on their own about the candidates.
As indicated above, focusing only on social media posts originating from the campaigns is just part of the story. And, as it turns out, it's a very small part, because social media buzz generated by posts originating from the campaigns accounts for much less than 10% of the total.
Throughout the primaries, I was monitoring social media buzz about the candidates. Two of the tools I used were Facebook's own "People Talking About This" data, which is presented on every Facebook Page, and USA Today's Facebook Barometer (which, as of May 31, is no longer updated). Facebook's "People Talking About" metric provides the number of likes, comments and shares a specific Page received over the past seven days. USA Today's Facebook Barometer measures seven days' worth of likes, comments and shares on the candidate's Page, PLUS the total number of posts across Facebook mentioning the candidate.
Comparing these two metrics at various stages throughout the campaign revealed that candidates' buzz generated by their own posts are a tiny fraction of their total buzz. In March 2016, for example, when Sanders and Clinton were generating around a million likes, comments and shares on their own Facebook Pages, their total Facebook buzz, including mentions originating on other people's Profiles, Pages and Groups was in the range of 24,000,000 to 34,000,000. This lopsided pattern was visible at all times throughout the campaign.
Given this tremendous discrepancy in buzz based on point of origin, anyone who analyzes the impact of social media on the elections without looking at the broader scope of the data has to be missing out on much of what is actually happening. We can't possibly know the full impact of social media on the elections UNLESS we look at the full scope of the data, not just the reactions to social media posts by the campaigns.
For further reading, here's a selection of articles written about the impact of social media:
The Power of Social Media in the 2016 Election by Gabrielle Gillespie
U.S. Election Buzz Tracker from Fleishman Hillard
Social Advocacy and Politics: Another Swipe at Social Media Engagement Predicting the GOP Primaries by Alan Rosenblatt, Ph.D.
Social Advocacy and Politics: Handicapping the 2016 Presidential Candidates with Facebook by Alan Rosenblatt, Ph.D.