In his hugely successful book Moneyball, Michael Lewis pitted people with data against people with feelings, and the data scientists won decisively. I am not entirely comfortable with this binary view of the world, because numbers acquire meaning through interpretation.
I propose a division of people into these two camps: those who have the data lead the story (data first), and those who have the story lead the data (story first).
Most of the pundits now commenting on the U.S. Democratic primaries belong to the story-first camp. They create a story to sell, and the round data are then fitted to their square script.
The story-first people reveal themselves most often by making arguments that sound plausible when taken individually, but tracing these arguments to their core assumptions, we'd find a contradiction.
Note that in a few states such as Iowa and Nevada, the winner is decided by caucuses. Caucuses are social gatherings in which the attendees decide the winner through a multi-stage process. Primaries are your typical elections in which you write your one preference on a ballot. The turnout at caucuses is predictably much lower that that in primaries, but the on-site process takes into account second choices, and also observers can learn more about why people voted for whom.
If you tuned into U.S. media in recent weeks, you won't find straight-up reporting of results (The [state] primary has concluded. [Candidate A] won with [x%] of the votes...). Instead, the newscasters are outdueling each other with the stories about the data. Let's take a closer look at two of the stories they came up with.
A) Bernie Sanders's blowout win in Nevada is deceiving because the small size of caucus-goers does not represent the will of the Nevadan people.
B) Joe Biden is the front-runner after his South Carolina victory because he now leads in popular votes despite being behind in delegates.
Underlying each statement is an opinion about the value of caucuses versus primaries. If a pundit pushes both statements, then this is not someone we can trust. Agreeing with one of these statements requires rejecting the other. Rejecting one of them requires accepting the other.
Start with statement A. Anyone who depreciates the Nevada results believes that caucuses are not representative (because of small sample size). The implication is that caucus results should be set aside or even ignored. (A statistician prefers to adjust the results rather than throwing out the baby with the bathwater. The adjustment is an attempt to bring the caucus results to a comparable basis with results from primaries.)
Now look at statement B. Anyone who says Biden leads in popular votes has no problem adding together caucus results and primary results. The implication is that this person thinks that the vote counts are representative of each state's population regardless of election format.
Therefore, if one accepts statement A, one has to reject B.
What if one rejects statement A, as a Sanders's supporter might? This person believes that caucus results are just as valid as primary results. This is basically the premise of statement B so this person should accept statement B.
Since one opinion (caucuses vs primaries) underlies both statements, we are not free to accept or reject these statements as if they are disconnected.
The election season provides many examples of story-first punditry. Various arguments around another divisive issue - one's attitude toward popular votes versus electoral votes in deciding election results. To put it simply, popular votes mean everyone's vote has equal value while electoral votes assigns different values to voters based on where they live and other factors. Famously, Hillary Clinton won more votes than Donald Trump in 2016 but lost the election because the winner is decided by electoral votes. It turns out electoral votes are also central to the Democratic primaries. Someone who is story-first decides which candidate they support and then find data that fit the narrative, resulting in flip-flops between arguments for and against popular votes.
News pundits have the luxury of controlling their airtime. If we have a chance to push back on them, the key is to draw attention to the issue underlying both statements. Do they believe caucus results can be compared to primary results? Do they believe in popular votes or electoral votes? If the answer is yes in one case, and no in the other, then we know we're dealing with someone who's fitting the round data into a square scripts.
My comments apply not only to politics but also to other settings such as business and sports. In the course of advocating a data-first approach, data scientists encounter many people with story-first tendencies. Effectively dealing with these interactions is a key to success.
P.S. Remember a few weeks ago, every pundit was screaming at us that the Sanders win in New Hampshire was a very bad sign for him because he did not earn 60 percent of the votes as he did in 2016. It's two weeks in a row now where the winner got 47-48% of the votes while the first-runner-up earned around 20%. Despite not reaching the 60-percent threshold by a wide stretch, these pundits have described both victories as landslides. They were wrong before, and they are right now.
Comments