Iowa Caucus – Can Social Media Analytics Predict Who Will Win?February 01, 2016
The Hypothesis & Methodology
I’ve been working with the team at Zoomph since last August. The challenge handed to me was to analyze social data to see if we can predict who will become the next President. Our hypothesis is social media data is a better indicator of election outcomes than polling.
As any pollster knows, there’s both an art and science to predictive analytics. There’s no way to be 100% accurate, but it’s fun to see how close you can get. For this project we’re using a consistent tracking methodology across each candidate whereby we’re listening for mentions of candidate as a “topic” of conversation on social media.
With a solid social listening methodology, I started bench marking national data each week to see trends versus the prior week’s data, and to a 28-day rolling average. We’ve seen correlation between this social media trending analysis and shifts in polls.
In the new year I’ve been analyzing state-level data to see how the race differs throughout the primaries. We identify location data by both the source of a post or via state-identification in an author’s bio.
With Iowan’s heading to Caucus today, rubber is about to hit the road.
Comparing Zoomph Social Analytics to The Des Moines Register / Bloomberg Poll.
On Saturday, The Des Moines Register (DMR) released the last poll before the big day. As noted by the team at fivethirtyeight, it’s historically been one of the most accurate polls.
For our comparison, I pulled social media data from 12:00 am January 29th to 11:59 pm January 30th, more recent than DMR’s telephone survey. I decided to look at only two of the many KPIs in Zoomph’s analytics suite, and then compare results to DMR. The KPIs I used: all social media activity with positive sentiment; and ZPoints® (an influence score) with positive sentiment. ZPoints is a proprietary algorithm that looks at several signals that measure influence of each and every post about a topic. For the elections I believe it’s a barometer that indicates more impactful and passionate conversations about candidates, and correlate it to those more likely to vote when comparing it to “activity.” I decided to only use positive sentiment too.
1. Donald leads by 1% over Ted using our method, compared to 5% in the Des Moines Register / Bloomberg poll (DMR).
2. Jeb is in fourth using ZPoints®, versus seventh in DMR.
3. Ben dropped 9% in our analysis.
Hilary leads by 3% using ZPoints®
There are many other metrics we could add to this single KPI approach I used for Iowa. Weighting additional factors may help refine and hone accuracy. But even with all of that considered, and as I’ve seen since I started this effort last August, there IS a correlation between Zoomph social analytics and poll results.
We don’t have to wait too long before we see how it compares to actual results. Rest assured, we’ll be glued to our computers watching real-time data streaming in from Iowa.