4 Reasons Why Social Media Analysis Won’t Accurately Predict Election Results — Yet
Lately the conversations around whether social media can help predict the outcomes of elections are popping up again. And rightfully so; the U.S. elections for presidential nominations are in full speed, and they’re the talk of the town across the world. As social media posts are a reflection of the talk of the town, whether news or opinion, these days a lot of buzz centers around politics and political campaigns.
So yes, we could say that social media analysis may help us indicate sentiment, and even predict who the winners and losers will be. Within this discussion, I’m missing nuance though. In many cases social media analysis simply isn’t complete or accurate enough to make predictions based on methods and models that prove to be effective in multiple cases.
Here’s four challenges we need to solve before we get to a level of maturity that will allow us accurately predict election results (partly) through social media analysis:
1. Social media posts reflect segments of the population
Not everyone is on social media, and in many analysis I’m missing the nuance as to segmentation of the layers of the population that make up a certain part of the sentiment. It’s too easy to say: Candidate A generates X% more volume than Candidate B, and only X% on Twitter is positive about Candidate A for example.
The challenge here is to get more accurate in terms of segmentation, and to try and make a link to ‘conversion’:
- Which segments do these people represent (e.g. demographics)?
- Are these people allowed to vote (age, residence, etc.)?
- Do they actually go out and vote?
2. Social media = extremes; nothing’s mediocre
Most of social media is about extremes; posts that are shared, are mostly about things or situations that trigger emotions and/or to get attention, raise discussion, etc. If someone thinks something is so-so, it’s often not worth posting anything about. Friends & followers won’t like, share or respond, and you’d like to avoid the ‘black hole’ effect where your social media post disappears without getting any attention at all.
All in all, people are more likely to look for the things they have a stronger opinion about, and we should take this into account when analyzing social media opinions.
3. Accuracy is still to mature
When we perform social media analysis, it’s never 100% accurate. There’s probably little research that is, but social media has to deal with a number of challenges, such as:
- Algorithms used for analysis and interpretation that are still learning and maturing every day
- Lacking or inaccurate information being processed
E.g. if someone is not a U.S. citizen, but posts something on Twitter from the U.S. (and includes location information), this may be included in a prediction — even if this person cannot vote.
- Private profiles of which the posts are automatically excluded, as they cannot be scanned (and therefore a part of the population is left out of the analysis)
- Generally lacking and inaccurate information and errors over large volumes of posts
E.g. social media posts without location information can be included in analysis, but may very well have no effect on actual voting when they’re not from a U.S. citizen.
4. Trolls and sarcasm can be tough to spot
Algorithms and Artificial Intelligence are getting stronger and more accurate by the day. Some social media analysis tools are language agnostic, and many are getting more accurate in spotting humor, trolls and sarcasm through contextual analysis. Clearly though, it remains a challenge even for people to differentiate e.g. sarcasm from positive opinion when it comes to sentences such as “Yeah, I’d sure vote for this guy!” or “Great job Trump!”. Without context and a better idea of people’s background, a look at other posts in their timelines, etc. it will remain hard to accurately categorize these posts in the positive of negative topic-buckets.
We’ll get there by 2020 though
Despite this, I must say I’m a strong believer of the added value of social media analysis. I’ve used it in my work for clients, and can be a great way to confirm assumptions or get new insights with regards to how people perceive your brand, actions, products, services, etc.
As social media analysis tools are maturing and the volume we can learn from is growing, I’m convinced we’ll get to more accurate predictions in the coming years. The upcoming 2016 US elections may come too soon, but I’d be curious to see how accurately we will be able to predict the Super Tuesday results in four years from now.
— — — — — — — — — — — — — — — — — — — — — — — — — — — —
Image source: Wallpapercave.com
Disclaimer: opinions, statements, predictions, and other content shared through social media by me in this message or other media are solely my own and not in any way supported by my employer.