Team Clinton worshipped at the altar and got burned.
Lots of post-election articles about how the Clinton campaign got fixated on their data-rich electorate models, using them to allocate ad dollars, deploy field workers and schedule “market visits” by Hillary and her surrogates.
What Team Clinton seemed to have forgotten is the old Reagan adage: trust but verify.
The data models – which worked near-flawlessly for Obama – took stage as “shiny objects” that led the Clinton campaign astray.
Politico reported a case study that illustrated the point …
Here are some snippets from a Politico article titled “How Clinton lost Michigan — and blew the election.”
The team on the ground around Detroit was desperate.
The Democrat’s analytical models projected a 5-point win through the morning of Election Day.
Michigan organizers were shocked.
Campaign headquarters was ignoring on-the-ground intel that the race felt like it was slipping away.
The people at campaign headquarters believed they were more experienced, which they were.
They believed they were smarter, which they weren’t.
They believed they had better information, which they didn’t.
The anecdotes are different but the narrative is the same across battlegrounds, where Democratic field operatives lament decisions drawn entirely from pre-selected data.
Headquarters just spit out “the model, the model”.
It guided decisions on field, television, everything else.
It was very surgical and corporate.
They had their model and were going to follow it.
On-the-ground operatives were told that their input was not “scientifically” significant.
When you don’t reach out to community folk and reach out to precinct campaigns and district organizations that know where the votes are, then you’re going to have problems.
The Clinton campaign dismissed in-person contacts.
So, no one was knocking on doors … hearing directly from voters … tracking how feelings about the race and the candidates were evolving.
This left no information to check the data and models against.
There was no early warning system that the race was turning against them in ways that their daily tracking polls weren’t picking up.
So, they missed that some of the white male union members they had expected to be likely Clinton voters were actually veering toward Trump.
People involved in the Michigan campaign still can’t understand why campaign officials in the Brooklyn headquarters stayed so sure of their numbers.
The models had predicted Clinton would beat Bernie Sanders in the Michigan primary.
He won by 1.5 points.
The models also predicted Clinton would win the Iowa caucuses by 6 percentage points.
She barely won … by with two-tenths of a point.
Still, their campaign logic was guided by full faith in their data.
When top aides to the Trump campaign mapped out the best-case scenarios for election night, they always fell short of 270, and Michigan was always the state that they couldn’t see a way through.
Still, Trump’s last stop of the election was a massive rally in Michigan that went on past midnight.
The Michigan rally was prompted by nervousness that theTrump’s campaign sensed coming out of Clinton headquarters in Brooklyn.
Walking out at the end, Trump turned to his running mate, Mike Pence and remarked: “This doesn’t feel like second place”
Democrats felt it too.
Rep. Debbie Dingell … has told people since the election that Hillary and Bill Clinton both said in their final appearances in Michigan that they felt something was off.
Still, on the morning of Election Day, internal Clinton campaign numbers had her winning Michigan by 5 points.
By 1 p.m., an aide on the ground called headquarters to alert that voter turnout in urban precincts was down 25 percent. Maybe they should get worried.
Nope, they were told: She was going to win by 5. The data said so.
But an hour-and-a-half after polls closed, Clinton aides began making rushed calls, redrawing paths to 270 through the single electoral vote in Maine and Nebraska.
Still assuming wins in Wisconsin and Pennsylvania, Michigan suddenly looked like the state that was going to decide the presidency.
An hour later, after they hung up, they realized it was over.
“They could tell by the numbers they were seeing — not the numbers being spewed from their own internal analytics team, but the numbers sitting at the bottom of the TV screen.”
A couple of teaching points from the Michigan case study:
1) Always trust but verify … balance the hard quant data with qualitative information that provides context & “texture” for the data.
2) The world changes … sometimes abruptly … sometimes too suddenly for the data and models to pick up the disruptions.
3) Always “triangulate” from multiple data sources … don’t let any single source become a “shiny object” … build statistical confidence by converging from multiple angles.