10.28 Presidential Social Intelligence Battleground Tracker — 7 Day Out Numbers
A pulse of how potential voters in battleground states are discussing the Presidential candidates
This post will take a different form as we near the end of our project. First, Happy Birthday to my wife.
Since 7/5 we have examined social intelligence and language/text data to better understand the presidential election in battleground states. (Please read the background post on this project if it is your first time reading.)
A full recap of our analysis, which is normally at the top of these posts, can be found at the end, along with a summary of some of the news coverage we have received in the mainstream media along the way.
You can also find all our normal explainer items at the end of this post.
Battleground States in our Analysis: AZ, CO, FL, IA, ME, MI, MN, NV, NH, NC, PA, TX, WI
Date Source: All data is publicly available and anonymized for our analysis from Twitter, Facebook, Online Blogs, & Message Boards.
Technology Partners: Eyesover & Relative Insight
Overview
To date, and in addition to other data, we have have aggregated, scored, and analyzed over 2.5 million Trump and Biden supporter social media posts that contain just shy of 60 million words.
This language dataset, along with candidate and topic/issue sentiment analysis, helps define our “Support Index Gap.”
Developed along with our technology partners, (Eyesover Technologies and Relative Insight) the “Support Index Gap” represents an analysis of the gap in measured support from battleground states between Biden and Trump based on social media posts from voters in those states.
Through analyzing and comparing the trends in the this gap over time we have attempted to establish a narrative around the election on a state-by-state basis.
“Support Index Gap” Summary Points
Simply, each state’s measurement design is built to a 50/50 scale for Trump and Biden (no undecided/third party candidates).
(Biden 52% v. Trump 48% would represent a postive 4% “Support Index Gap” for Biden.)
— This approach is meant to show trends more than exact ballot position. It is not designed to compete with polling, but complement it with more real-time context.
— It measures the gap in support vs. the candidates based on the online conversation (which is subject to big swings) and is not balanced to turnout models.
— However, since we looks at so much volume, (tens of thousands of posts a day) in each state on a daily basis, it helps normalizes the data.
— We have observed many instances where these numbers tighten and expand in both positive and negative moments for each candidate.
— There are random outliers, like Iowa, where the support index is clearly out of whack from what the ballot will likely be. Does that overwhelming trend prove an outlier in level in actual support? We will see.
— In this approach it will also be very valuable to look back on the trendlines once we know the final results to see where things were wrong/right, or the data was more predictive than we noticed at the time.
— This is the first time we are measuring a Presidential Election in this manner — we will learn more looking back after the dust settles. For instance, perhaps we will be able identify a slant (or margin of error) in our data that we can apply in the future.
— Ultimately, campaigns are more about turnout than convincing/persuading undecided voters on the margins. If Trump is successful in surgically finding new voters in key states it will certainly counteract macro national trends from an electoral college perspective — if Biden, it will extrapolate those trends. We won’t know until the polls close.
The Data — Support Index Gap
The “Support Index Gap” shows Biden ahead in every state, therefore, the numbers have been presented in terms of Biden’s positive support gap. Any number (unless it is negative, which would mean the gap flipped to Trump) represent a gap in favor of Biden.
(4, +4 or 4% would mean a 52 to 48 measurement of a Biden favorable Support Index Gap)
We have presented the data in three grouping:
— the day-by-day gap trendline from 7.5–10.27
— 7 day gap averages compared week-by-week from September and so far into October.
— September and October monthly averages
7 Day Week-by-Week Averages
September and October Monthly Averages
Lots of Graphs
7 Day Week-By-Week Averages:
September & October Month Averages:
7.5–10.27 Support Index Gap, Day-By-Day Trendlines:
Project Background
Support Gain/Lost Overview
The “Support Gain/Lost” metric is built by analyzing the language used in social posts about the presidential candidates. As battleground voters use language to indicate support for a candidate, our systems register “+1” in “Support Gained” for that date. A post expressing negative support is registered as “-1” in “Support Lost.”
These numbers are based on holistic expressions of support gained/lost, or shifts to neutral, from a unique social account. Only shifts in categorization are registered per social account after an initial signal is measured. (EX: a post from someone supporting Trump is not counted two days in a row.)
There are 6 metrics we monitor on a daily basis across each state:
— Previous Non-Supporter to Neutral: (a post indicating a previous non-supporter has moved to a more neutral stance on a candidate)
— Suport Gained: (a post indicating support gained for Trump or Biden)
— Support Lost: (a post indicating support lost for Trump or Biden)
— Previous Supporter to Neutral: (a post indicating a previous supporter has moved to a more neutral stance on a candidate)
— Net Daily Gain/Lost: (the daily net of Support Gained v. Support Lost)
— Net Daily Neutral Shift: (the daily net of “Shifts to Neutral”)
Aggregate counts from July 5th through October 10th:
— Trump Total Support Gained: 195,301
— Trump Total Support Lost: -143,473
— Trump Support Net: 51,828
— Biden Total Support Gained: 126,472
— Biden Total Support Lost: -113,080
— Biden Support Net: 13,415
Language Analysis
Building off the “Support Gained v. Lost” classification, we are aggregating an anonymized dataset of social posts that represent support for each candidate. This dataset allows us to compare and analyze the language used by “Biden Supporters” v. “Trump Supporters.”
From this analysis, we can better understand the issues/topics/trends that cause voters to express support or opposition toward a candidate. It also exposes overall patterns from the language used between the two supporter datasets.
— The analysis is processed via the Relative Insight language comparison platform.
The chart below shows the current size of the language data set:
— 1,188,264 Trump Supporter Posts
— 26,701,654 Trump Supporter Words
— 1,398,264 Biden Supporter Posts
— 31,959,495 Biden Supporter Words
Recent WSJ Coverage
Our work on this ongoing analysis was recently featured in The Wall Street Journal:
Election 2020 Polls: Startups Pitch Themselves as Alternatives
Many campaigns, political groups still pay for polling, but use online surveys and social-media analysis to understand voters
Recent Coverage in The New Yorker
Can We Trust the Polls?
Polls are not predictive; they are snapshots taken at a particular moment in time — one that will have passed by the time a poll reaches the public.
Past posts:
— Our initial dataset from the week of 7.3 was detailed here.
— Data from the week of 7.11 was detailed here
— Data from the week of 7.19 was detailed here.
— Aggregate data to date through 8.14 was detailed here
— Language data from the weeks of 8.07 & 8.14 was detailed here
— Aggregate data through 8.25 was detailed here.
— Language data from 8.21–8.28 was detailed here
— Language data from 8.28–9.04 was detailed here
— Aggregate data through 9.5 was detailed here
— Language data from 9.04–9.11 was detailed here
— Aggregate daily support gained/lost data through 9.17 was detailed here
— A first look at Supreme Court Data was detailed here
— A deeper dive into Supreme Court Data was detailed here
— A language comparison between voters in NC and FL was detailed here
— An overview of the pre-debate top issues was detailed here
— Aggregate daily support gained/lost data through 9.30 was detailed here
— Post debate issues and sentiment were detailed here
— Trump’s COVID diagnosis and recent campaign issues flow was detailed here
— A language analysis of data from 9.25–10.2 was detailed here
— Aggregate data through 10.11 was detailed here
— Post second debate data was detailed here
About The Author
Adam Meldrum is an award-winning political & digital strategist. He is the Founder/President of the Republican media-buying operation AdVictory LLC. Adam has advised other organizations in the political/technology space such as DDC Public Affairs and WinRed. Adam also serves on The Board of Directors at HeadCount, a non-partisan organization that uses the power of music to register voters and promote participation in democracy.
Since 2006 Adam has produced award-winning work for and advised campaigns/organizations such as; Governor Rick Snyder, Governor Bruce Rauner, Governor Doug Burgum, Governor Bill Lee, Ambassador Ron Weiser, Senator Ben Sasse, Senator John McCain, Senator Rand Paul, The Republican National Committee and the U.S. Chamber of Commerce.
Adam’s work on the forefront of technology, data, and analytics was featured by OZY Media in a profile: “Meet the GOP’s Chatbot and Artificial Intelligence Guru”. He is a die-hard Michigan State University fan/alumni and pretty obsessed with Phish and The Grateful Dead. He resides in Washington D.C. with his beautiful wife Christina and their weird puppy, Basil.
Personal Website: http://www.adammeldrum.me/
