Today, we'll look at PoliticIT.
PoliticIT.com is a site that measures candidates "IT" scores based on their online footprint. Things like Facebook and Twitter followers, Wikipedia checks, and (ugh) Klout scores combine through proprietary algorithms to produce the final IT score. Their claim: the guy with the highest IT score wins 87% of the time.
|Barack Obama: 87% Likely To Win|
What's The Deal?
It appears they want to vend "big data" predictive services to clients (presumably political campaigns). From a twitter conversation they say that ranking for the various metrics is ordinal rather than cardinal (so it's More vs. Less) and that the weights are determined via a neural network assessment (so does Klout score count more than Wikipedia updates? The neural net figures that out ... I guess).
What Does It Mean?
It's hard to know: they are new. Their website includes no bios for the team, their "footer links" are not links and do not go anywhere. They do not make available track-record, etc. So I will speculate and invite them--they are active on Twitter and responsive--to talk to us about what they're actually doing.
Here's What I Assume
Neural nets require, most of all, feedback in order to be able to learn. I presume that there are two kinds of feedback they are pumping into their system: polls and actual election results. I would further assume that all the raw data (InTrade scores, etc.) are time-stamped so that they can measure correlations between inputs (does Twitter go up the same time Facebook followers do?).
This should present a relationship between network events (InTrade price spikes) and feedback (poll numbers climb). If they can manage that carefully enough they could maybe tell a candidate what a bunch of hits to their Wikipedia page mean (it means they're going to be Romney's VP, obviously).
The Big Question: Demographics
The piece of the puzzle that I do not see is the most basic one: demographic data. Just as it's clear that updating your own Wikipedia page a whole lot won't make you Romney's VP, understanding that a lot of Twitter followers means you are "more likely to win" (so if the whole population of China follows me ... I'm a lock? Obviously not) has to actually map to a real-world behavior pattern by real people (i.e. why and who are all those people following me? Are they voters?) for it to be actionable.
In order for a candidate to "have a lever to pull" to make things happen they have to understand what is behind the pattern of behavior the Internet is showing. That's hard: Wikipedia and InTrade don't (as far as I know) collect / publish demo-data. Facebook's a gold mine. I don't know about Twitter. Klout is at least one-step removed (their score is a combination + proprietary algorithm of some of the same things that may be monitored separately) and so on.
So if I don't know who is doing google searches or looking at my wikipedia page ... how do I know what action to take to get more voters?
The Second Question: Correlations and Causations
One of the promises of Big Data is that it can (at least in theory) "know things" because of a storm of aggregate information points that are simply not clear and not intuitive from looking at human-sized data-blocs. In other words: we may not know why having a lot of Twitter Followers outside the country means you are more likely to be elected inside the country--but if the data, properly analyzed shows it's true then it "it's true."
However, this analysis becomes much, much harder when the variables being tracked are related to each other. If Facebook followers moves in some degree of lock-step with Twitter followers then we have a very murky view of cause and effect that makes trying to get insight out a lot harder (this goes double if I am "buying" Twitter followers which, at least some people, have accused the Romney campaign of doing).
I suspect that these data sources are, today, so poorly understood that PoliticIT has a massive uphill battle to fight. I'm very curious to see what marketing material they bring to the table in order to convince a potential buyer that they have real actionable insight. I have only one guess as to what is--and this is nowhere on the site ...
Possible Value Proposition: Event To Effect Matching?
The blockbuster golden value proposition that I think they could have is this: what if they propose to be able to show, in real time how events in real life correlate to variance in their score faster than any reasonable polling cycle. That is: Obama releases his (widely vilified) "Understands" video, we see articles decrying it--and want to know "what's the effect"? Well, let's see what happened with PoliticIT's competition the other-other big data Internet poll: the Twindex!
|Bang! That's a ski slope!|
If the only thing in the universe that happened in the campaign universe was that one video (which is considered, pretty much, the dirtiest video yet slung) we could safely assume that caused the drop. Maybe it did.
But that wasn't the only thing that happened in the universe--there was a storm of other things--and even in "campaign space" there was a massive number of articles published, ads seen, and speeches given. How fast a given piece of data propagates is a hugely complex issue (see ad agencies: everyone "wants to go viral" but no one "knows how to make that happen reliably"). So we don't even know if people are reacting to something that happened the day the decline hit ... or before then ... or a number --a critical mass--of things--or whatever.
If PoliticIT's geniuses (quants) can sort that out--can link individual movements to different demographics to different trigger events (i.e. "What real-world event makes your mom check wikipedia?")--then, well, they have something. If it's accurate--and coupled with 'what it means to your success chances'--then it's a license to print money.
But the above is a total guess: they have a great, high-tech web-site but nothing on there that I saw suggests they can or are even trying to do this.
PoliticIT guys: What say you?