SoylentNews is people
SoylentNews
Submitted via IRC for chromas
The Clinton presidential campaign used a complex computer algorithm called Ada to assist in many of the most important decisions during the race.
According to aides, a raft of polling numbers, public and private, were fed into the algorithm, as well as ground-level voter data meticulously collected by the campaign. Once early voting began, those numbers were factored in, too.
What Ada did, based on all that data, aides said, was run 400,000 simulations a day of what the race against Trump might look like. A report that was spit out would give campaign manager Robby Mook and others a detailed picture of which battleground states were most likely to tip the race in one direction or another — and guide decisions about where to spend time and deploy resources.
Of course, the results are only as good as the data. Since the outcome of the election was different than most poll predictions, it seems like Ada may have had a Garbage In, Garbage Out problem.
(Score: 1) by khallow on Sunday November 20 2016, @02:42PM
More likely the algorithm assumed that the Sanders supporters wouldn't refuse to vote for her in such large numbers. And it almost certainly wouldn't have factored in the possibility that a lot of them would wind up voting for Trump. Nor is it likely that the algorithm would consider the large turnout amongst rural voters that hadn't really been showing up to vote in recent years.
Now the algorithm and data are black boxes to us. We don't know what the algorithm considered or whether and how often the Clinton campaign polled for the relevant data above. But it is worth noting that such gaps in predictions are not hard to characterize logically in a situation with limited choices.
For example, a potential vote by an eligible voter has three possible outcomes: vote for Clinton, vote for some other presidential candidate who is not Clinton, and no countable vote (say by messing up their ballot or not voting at all). You can then further split out the various classes to attempt to capture behavior or outcomes of note. Same goes for characterization of voters' preferences (such as Bernie Sanders voters in the primaries), voter context (such as rural voters), etc.
I don't see any obvious missteps of the Clinton campaign versus any of these categories except perhaps in their heavy handed attack of the FBI Director Comey's letter at the end which I doubt was steered by algorithm or perhaps some discounting of rural groups and such in swing states (Ohio, Pennsylvania, and Florida) which might have simply been something that they could do much about at the time even if they did know there was a problem. It appears to me that for the most part, they had good data and they used it well.
The problem as I see it, is that Ada was meant to be a considerable advantage and appears to have successfully served as such. But no matter how great your advantages, if you're in a sufficiently bad losing position, you will lose (unless you can metagame yourself out of the situation, of course, which wasn't an option available to Ada). In the end, Ada couldn't compensate enough for the quality of the candidate.
That's really worth noting here. Clinton outspent Trump by something like half a billion dollars, had this sophisticated algorithm on her side supposedly, and had run a much better organized and politically supported campaign. Yet she lost. We aren't going to find the cause of her loss in some algorithm or in fake news on Facebook or Google.