paint-brush
AI & the Midterm Electionsby@crypto
538 reads
538 reads

AI & the Midterm Elections

by Percy Venegas EconomyMonitorOctober 15th, 2018
Read on Terminal Reader
Read this story w/o Javascript
tldt arrow

Too Long; Didn't Read

AI is larger than machine learning. The physical world around us and our society compute themselves, constantly. What is the use of glorified statistical models that tell us something about how the world was yesterday -what happened in the last election- if our machines do not perceive what is important to humans today, the sudden & subtle changes in human priorities?

People Mentioned

Mention Thumbnail

Companies Mentioned

Mention Thumbnail
Mention Thumbnail
featured image - AI & the Midterm Elections
Percy Venegas EconomyMonitor HackerNoon profile picture

Can AI improve midterms forecasts?

Context awareness

AI is larger than machine learning. The physical world around us and our society compute themselves, constantly. What is the use of glorified statistical models that tell us something about how the world was yesterday -what happened in the last election- if our machines do not perceive what is important to humans today, the sudden & subtle changes in human priorities?

People vote with ballots and with their wallets

People really care when they put money on the table -it is unlikely that a voter who was materially committed to a cause by means of a political donation does not show up to vote. That measurement of trust is a signal (a driver) that machines can count on. But is also possible that after outspending on amplification, the contest is lost because resources were misallocated. Luckily, once provided with purpose, machines excel at planning. Therefore, political campaigns need to understand those trust signals and optimization objectives.

Online contribution platforms

We study the user behavior in online contribution platforms, to discover what campaigns are more/less successful -then we study those campaigns individually to learn why they are successful.

Since this is a quantitative macro approach based on open data, we do not access individual private data at any point — we are concerned only about group behavior and observed probabilities (what actually happened). The data is generated for specific periods of time (e.g any month going back 2 years), or on what is a nearly real time in the realm of campaign finance (2 days ago to 4 trailing weeks). Weighing for recency allow us to account for the effect of momentum.

MyNGP

One good example to look at is the NGP platform.

The popularity of campaign contributions

The figure shows thousand of campaigns and movements that collect donations. Some may capture a large share of attention at a moment in time, but do not sustain grow. Some are evergreen: they generate growth outbursts many months after they first started. Most are too little to likely make any long-lasting difference. And there are the whales: the candidates or causes that account for dozens of the small ones. But how?

You can find all sort of surprises. For instance, democrats use republicans overtures to advance their own narratives, like in the case of one of the top donation pages in Actblue (Actblue is another fundraising platform frequented by democrats) that is not about a democrat but about a cause, and actually uses the position of a former republican presidential candidate as a pitch for “tax fairness”.

Other services, such as Anedot, are popular among GOP candidates and are worth studying to develop a digital campaign strategy that covers multiple angles.

By having this information campaigns can compare their performance to the most successful within their own party, but also between competing political parties — then, replicate the tactics that work.

Augmenting human intelligence

Often times, traditional methods such as polling and history of fundraising are not definitive — a few seats are too close to call. It can even happen that “expert” forecasts and comparable-based statistical methods are completely off. In those cases the embedded intelligence of AIs that sense and compute the belief consensus -of people who actually pay attention and commit resources- is akin to a professional punter.

Because even Nate Silver FiveThirtyEight has to toss-up a coin, sometimes.

Fivethirtyeight, ABC

The rise of decentralization in politics

One of the main takeaways is to see the very low traffic in the majority of the small district campaigns — they have no forceful online media strategy.

In general, pages where the candidates branding is strong and the value proposition/message/imaging is loud & clear, perform well. Pages that simply ask for money upfront do less well.

A bad way to ask for money, is to do nothing more than asking for money

But there is something more fundamental going on: the decentralization of political financing. One can see it with the prevalence of peer-to-peer (P2P) texting; the bitcoin community rally behind libertarians, greens, and Sanders during the past election; the emergence of mobile Apps as social media alternatives to convey a discourse and to build grassroots movements.

A political risk AI is particularly good at doing cohort analysis that shows the “churn” patterns of the users of popular political apps. With these insights you can understand when you will “lose” the attention of your voter, and when the re-engagement activities should be programmed. For instance,

1.You can capture the attention of voters that are interested in your competitor.

2.You can re-capture the attention of someone who visited your fundraising platform, even if they did not committed funds.

Having this information on hand, operational implementation is straightforward: campaigns can engage person to person in the places where THEY ARE, or use programmatic marketing services to display ads in those places specifically (e.g. where & when they consume news). Without this knowledge, you will be shooting in the dark, and burning promotional budget.

Case in point: A candidate that did not win the primary in his congressional district, but received 80,000 online donations. Desktop users of his website on a particular month were mostly male (51.84%), and over 65 years of age (22.61%). In the other hand, mobile users were predominantly female (59.2%) and between 55–64 years old (22.84%). Younger generations (millennial, etc) were a small fraction in both cases. This difference matters because that month the website received 72.89% of visits from mobile devices, which indicates that there was a “blind-spot” in terms of reach — or perhaps an opportunity to engage that population (around 19.1 thousand women) as standard bearers.

The candidate had a strong personal brand, 60.44% of the visits from search engines such like as Google came from the candidate’s name query (around 4,630 visits), and it would cost him $ 0.34 per click to get those visits using paid ads. Another sign of his strong candidate brand is the fact that 81.14% of the visits to the website were direct (not coming from email, or search), and more than 70% of the visitors checked more than 1 page before leaving the site.

In his case, fundraising was only half way through: he needed the intelligence data to understand where his political efforts were at risk.

The digital realm is a competition for attention — if people are interested in local politics, but are not spending time at your site or at your competitors, they are probably hanging out where the undecided are. And today, that happens to be mobile, the handful of apps that people use repeatedly, and their cross-app usage in the intersection of life and politics.

Based on a political risk forecasting algorithm developed by https://www.economymonitor.com

EM’s Percy Venegas is also the author of Trust Asymmetry, published by Springer.