How I saw the Brexit outcome two days prior using an app.
The pre-Brexit world of 2015 looked a lot different even though it was a mere five years ago. Fringe opinions on politics and society were still fringe. Most people felt reluctant to express what they really thought about contested topics like immigration, gay marriage and welfare. Such debates took place under the veil of anonymity on platforms and forums in which bullying and hatred were common themes. It wasn’t rare for someone to express their opinions on these platforms only to be met with a barrage of insults. GDPR wasn’t yet a thing. The comprehension of ubiquitous data collection and processing was reserved for only those in the field or those in academia. This meant most users of technology platforms were relaxed about providing sensitive data that they wouldn’t necessarily be comfortable with today.
It was around late 2015 when two friends and I began thinking hard about the issues of debating and opinions. We often found that the only platforms where people went to debate were usually forums and social media platforms. Apart from the issues mentioned above that these platforms had, they were very noisy. Almost all social media platforms had different verticals and topics competing for real estate on your screen. For example, in a single news feed, you could see a video, a photo and a status update from your friends, news outlets, celebrities and the Pope. This meant that almost by design, debate or the exchange of opinions was restricted to the congested comment section of those posts.
The biggest issue by far was the linkage to your public profile; it was easy for anyone to see who you were and your historical activity. Now, this wasn’t a flaw of these platforms, however, it makes one think twice before engaging naturally with a piece of content. It was only when someone’s identity was hidden, that they felt more comfortable in expressing what they felt. Unfortunately, it also meant that a hidden identity made you more relaxed with generating hateful and unnecessary content. Another issue we saw was the overwhelming development of news stories that occurred and rendered at any given time; it was and still is, hard for people to keep up with events around the world.
After several weeks of brainstorming and mediocre anecdotes, we decided to develop an iOS and Android application with a very simple concept; every day at 07:00 am we would ask you one single (current affairs) question that was relevant that day. After voting on the question, a forum would appear where users could engage with one another. For example, a user would see the question Should the bedroom tax be scrapped? to which the user could vote Yes or No. After their vote is cast, the forum would open where they could see the comments of other users. There wasn’t a real end goal or a commercial plan for the application. Instead, we just wanted to build something that was community-based, safe and meaningful to its users. The timing of such an application was precise. With development almost complete across both platforms, the Brexit referendum was announced only several months after Donald Trump declared his presidential bid. In addition to this, things were heating up geopolitically from Russia-Ukraine, Le Pen and the Middle East to Black Lives Matter. Suddenly, everyone had an opinion.
In order for us to have really gauged opinions on different topics, we had to have had granular socio-economic data about our users otherwise the responses would have been useless. At the time, we didn’t want to build intricate and obscure logic to gather this data in the background so instead, we opted for a more simple approach: we just asked users for their data. Upon signing up, we asked users for detailed information about themselves that today would be unthinkable. We asked users for data about their location, education level, party affiliation, race, income and more. Most importantly, we hid the user’s identity on the application by asking them to create a username for others to see. We figured that if users were anonymous, they would feel much more comfortable providing data and expressing their views. In addition, this reduced the capacity for users to engage in nefarious behaviour as it didn’t make sense to attack some you knew almost nothing about: no profile picture to judge, no tag lines to scrutinise and no past activity to stalk. Just a username.
With development complete and the world accelerating towards the pinnacle of polarisation, we launched the application with little fanfare. Very few people were signing-up and we weren’t getting much engagement. Users weren’t doing much after voting: there were no debates taking place, just hazy comments. Again, we thought about what we could do to break the echo chambers people were so accustomed to. We wanted to get different people into the same room. Like any good marketers, we decided to run micro-targeted advertisements on social media platforms and other websites to promote the application to different types of audiences. We ran ads based on location, age, hobbies and more with each ad using different messaging and banners. It worked really well. A few months down the line our application had amassed a userbase in the high thousands with sign-ups across the country. We also achieved some word-of-mouth growth from incumbent users telling friends and people they knew. In short, we accumulated a fairly representative sample of the UK population and enough for us to infer.
With this, we were able to finally get real, organic conversations going on a range of topics we asked about. The results were hard to fathom: strangers voting and having civil debates on the internet. Magic. A powerful aspect of the application was the very low nonresponse bias we had achieved. This was down to the low friction involved in responding to a question: one tap was all it took for a response to be submitted and recorded by a user upon opening the application. This ensured a very high user response as soon as our push notifications went out at 07:00 am. That being, not all users engaged in debate.
As the UK slowly approached the referendum day, we began to ask about topics that were being pushed by both the Leave and Remain politicians. This included the NHS, immigration, sovereignty and more. As we continued to ask, we started to randomly partition users into subpopulations to see whether outcomes differed much. In addition, we also looked at how users with different socioeconomic factors responded to certain questions. It is worth noting, however, that we didn’t factor in some bias that could have been part of the polling we did whether it was the wording of our questions or perhaps not having a fully representative sample as we thought we did. Nonetheless, the data showed consistency in the way users were responding as we continued to try to reduce sampling errors by either acquiring more users or by trying new partitioning methods. The results we were seeing from our userbase showed that there was a division between leave and remain answers with some outcomes favouring leave and some remain. This could be down to the different response rates we had for a given day which skewed some outcomes.
Two days before the day of the vote, we flat out asked users whether they were voting leave or remain. The question resulted in most users voting leave with roughly 60% of total votes opting to leave. Of course, we were sceptical at first as we thought users would vote on what was deemed the most radical outcome. However, we decided to look at how both users voted on past Brexit-related questions. The questions we asked leading up to this one suggested that the users who voted to leave were responding to previous questions that favoured leaving. Users showed distaste for things like EU-related taxes, EU-enforced immigration policies and of course the infamous £350m a week to the EU. More interestingly, a substantial proportion of users who voted in favour of remain topics still voted to leave. Perhaps some users agreed with some remain arguments but still wanted to leave the EU. We then drilled down a little further to find on average there was consistency in votes across different characteristics found in users and concluded that users, for the large part, did vote truthfully. Sure it was slightly nuanced but for the first time, we had a realistic simulation of what the outcome could be.
The forum for the leave or remain question revealed more interesting information and just how divided users were, something that mainstream polls lack. For example, some users admitted voting against their preferred political party’s position. Unsurprisingly, the common themes around Brexit were being debated with users becoming heated for the first time. That was the first time we realised that outcome of the referendum vote could, in fact, be leave, which was still seen as unrealistic back then.
Several days later, the referendum result was announced and people across the country were expressing what they really thought. Although we saw it coming, there was still a sense of disbelief that the outcome was what it was. As for the application, we decided to stop development and close shop after a year of more questions and disposed of all the data we had. This was mostly due to time constraints we had elsewhere and again, not having a clear mission for the application.
Looking back today, it felt like we had gazed into a pandora’s box of the future, except today those disagreements are far more forceful and violent and are taking place outside the half-baked app we had built. What went on to happen of course is now history. The world has become more authoritarian, nationalistic, xenophobic, unilateralist, anti-establishment, anti-expertise and populist. There are of course many opinions as to why this is. Technology and the science of data cannot constantly be framed as the problem or perpetrator. That would be far too easy and dangerous to conclude. Instead, we must responsibly employ technology and data as tools of progress and development whilst continuing to understand its potential and limitations. Of course, one can point to several instances where the role of technology has contributed to harmful outcomes.
Nonetheless, we must continue to create more dialogue with all of those in society so that all members are able to freely and safely express their opinions and views, regardless of how they look or what they believe. The world is big enough for everyone.
Ziad.
