You’ve heard of wardrobe malfunctions. But how about data dysfunctions?
As my colleague and MediaPost Editor in Chief Joe Mandese put it, the big takeaway from MediaPost’s
Marketing Politics opening panel yesterday was that pollsters made assumptions based on data from the 2012 presidential election.
And that was just flat-out wrong and amounted to a
“great data dysfunction,” according to Mandese’s report.
“Taking likely voter samples from 2012, but not weighting or adjusting, based on current trending,” was the chief mistake
most major polling organizations made,” said Matt Oczkowski, head of product at Cambridge Analytica. “If you had based your weighting and sampling based on 2012, you’d have a far
different view of what happened this year,” he added.
And Tyler Brown, director of digital operations at the Republican National Committee, noted an over-reliance on “predictive
analytics” versus “prescriptive analytics.” He said the role of data analytics should not be “you’re going to open a door and see the future,” but to use it as an
indicator to determine “where the holes are.”
Perhaps the data-heavy programmatic media sector should take note of these takeaways from the 2016 presidential election. For a sector
that relies on data — first- and third-party and the integration of that data — perhaps there are lessons to be learned. Maybe the industry should not rely so heavily on algorithms and
predictive analytics for every single thing. Those target audiences marketers talk about — they’re us: you, me, and everyone we know. Real people. We’re not
“targets.”
Go out and listen to us...