Barack Obama did not win the election because more Americans thought he would be a better president than Mitt Romney. More Americans voted for the incumbent than for the challenger, but it is Obama’s superior campaign organisation, and not his personal appeal, that deserves most of the credit. In particular, his product managers were better than Romney’s at using the technique of “data mining”.
The technique, pioneered by supermarkets, is conceptually simple: measure everything and tweak as necessary. In practice, it is a delicate affair. Suppose a popular soft drink has 4 percent higher sales when it is stocked next to a salty snack than when healthier raisins are its shelf-neighbour. Should shelf locations be swapped? There are many variables: the effect on sales of salty snacks and raisins, the profit margins of the different products, and customers’ sensitivity to any price changes. Most of the effects are tiny, but the study of millions of data, including a large number of computer simulations, can increase a retailer’s revenue and profit by a few percent.
In elections, data mining can bring votes to candidates and can increase the supply of contributions which pay for vote-gaining advertising. The work is detailed. Time magazine reports that the Obama campaign carefully tested how much more likely undecided voters in each close state were to yield to the blandishments of local rather than to out-of-state volunteers. The superiority in detailed computer work – “We ran the election 66,000 times every night”, as one expert explained to Time – probably gave Obama a few more percentage points of votes than Romney. It was the margin of victory.
Is data mining good, bad or ethically neutral? Supermarket executives may say that the practice is good because it increases profit, but profit should not be the only goal of any company. A more nuanced judgment is appropriate. The careful study of purchase patterns is truly valuable insofar as it helps consumers shop more wisely and helps stores reduce wasteful investment in inventory. It is pernicious insofar as it manipulates consumers into buying things they do not or should not really want. By this higher standard, commercial data mining is at best a mixed blessing.
In politics, political data mining would clearly be a good thing if votes were all that mattered. In the terminology of marketing, this scientific electioneering is a proven, cost-efficient method of improving electoral outcomes. However, an election is supposed to be quite a different matter from a contested retail market or a sporting event. The candidates’ goal should not be to win at any cost, but to help voters shape the nation’s future by giving them a clear choice of policies and philosophy.