Sunday, February 21, 2016

Do three people know less than one (about Nevada's Latino Voters)

While reading this, spend a minute listening to one of the greatest Car Talk talk bits.

It will explain how bad political memes (and other dumb ideas) happen.

If you want to really get into that, here is the letter mentioned in the above one minute bit.

Since I am about to tread on sacred ground, let me first hedge by saying that on average when Ezra Klein, Nate Silver and Nate Cohn agree on something it is more likely to be accurate than 3 randomly chosen pundits. Sometimes by a wide margin.

But their notion about Bernie Sanders not winning the Latino vote in NV started with one of them not knowing what he was talking about and then another (cue the Car Talk bit above) and ....

It started when Nate Cohn wrote too much with too little time to research over 2 hours of tweets during NV election results day using simple back-of-a-napkin-sized math.

This nonsense was then blessed by the other two (along with an article of
IF you assume A, B, C, D, E, F...
THEN our theory is correct).

To go point by point on that article to detail in depth exactly why each of its assumptions are suspect is beyond the scope of this blog post (and my currently available time).

That article supposedly points to this NY Times analysis of Nate's, but the included URL just goes to a page showing the actual total delegate counts at the time. So I have included all the relevant tweets in this.

That other well known analysts in that tweet conversation like pollster/poll analyst Mark Blumenthal and Harry Enten and a few unknown data scientists (but one data scientist worked on the '12 Obama team) were skeptical, is never addressed by the trio (or even mentioned in the Vox article above). And since Ezra and the Nates carry such weight in the political pundocracy, you can count on this bad idea spreading like wildfire - starting with the Sunday talk shows and then really raging by Morning Joe early Monday.

I'm another unknown data scientist, but I've spent 3 decades analyzing/forecasting data and am currently working towards a master's at Harvard in my field to get some name-brand credentials. But credentials shouldn't matter. Common sense should suffice.

I'll end with one last ancient pearl of wisdom:

Occam's Razor

Among competing hypotheses, the one with the fewest assumptions should be selected.

A bunch of people (entrance pollsters) tried to randomly select voters at caucuses.
The lengthy arguments in the Vox article and Nate Cohn's Tweets

You decide which has the fewest assumptions.

And then if you still feel that Latino's can't possibly favor Bernie Sanders, please explain how the complex set of motivations all kinds of different people have when voting, can be boiled down to
  1. what race you are 
  2. that a Senator spent most of his life not living near people of your race (neither did the other Senator, but I digress)
And please do so with the smallest number of credible assumptions.

P.S. Yes I support Bernie Sanders, but I don't support bad math, bad statistics or bad data science.

No comments:

Post a Comment