Possible Lessons fron the NY-26 Special Election

[Arwon]

As a sample, an entire electoral district's worth of voters is a pretty robust sampling of opinion, much better than any poll. The margin of error will be tiny. It in fact represents a census of voters in that electorate (ignoring the turnout issue in countries without compulsory voting).

What BC means to say is that the sample is not representative of voters in a nationwide presidential election. And he's right too. It's biased systematically by the demographics of the electorate (although we can make inferences about similar demographics elsewhere), by the circumstances of the vote, and by the turnout.

The turnout rate alone (sub-50% here versus, what, 65% or more for a presidential election?) means the sample is systematically biased in certain key ways (my guess: this turnout was on average older, richer, whiter and more partisan than the average of voters who turn out in a presidential race).

 

[Zelig]

What BC means to say is that the sample is not representative of voters in a nationwide presidential election.

No, I really don't think that's what he means, because he keeps talking about a sample size of 1 not being statistically significant.

 

[Arwon]

Ah.

Wouldn't making the entire district the unit of analysis require the assumption that each district (each an assemblage of thousands of individuals) is a single discrete data point, and that each district is independent of every other?

 

[BasketCase]

What BC means to say is that the sample is not representative of voters in a nationwide presidential election. And he's right too. It's biased systematically by the demographics of the electorate (although we can make inferences about similar demographics elsewhere), by the circumstances of the vote, and by the turnout.

No, I really don't think that's what he means,

WRONG, Zelig. Arwon got it exactly right.

 

[Arwon]

It's not a small sample size though.

 

[Zelig]

WRONG, Zelig. Arwon got it exactly right.

Yes, Arwon got it right, but what he posted is not what you communicated.

 

[downtown]

Ah.

Wouldn't making the entire district the unit of analysis require the assumption that each district (each an assemblage of thousands of individuals) is a single discrete data point, and that each district is independent of every other?

Yup! NY-26 isn't one data point, but tens of thousand!

I've read dozens of academic papers on US congressional elections (it was one of the two concentrations in my degree, the other being ed policy) that referenced less than 10 races at a time. There are some very smart people making a hell of a lot of money extrapolating data from NY-26 to be used nationally as well.

Clearly there are some specific circumstances here that we can't apply to other races. The interesting there here isn't so much that a Dem won, its why the GOP candidate performed so poorly, given the demographics of the race. I don't think Congressional Democrats are hoping to seriously compete in too many R + 5 and above districts, but if they can apply some of the reasons for the underperformance here elsewhere, they won't have to.

 

[BasketCase]

No, I'm not letting this go. Was simply offline for a few days while I installed some new stuff in my PC case......errr, just a second, I'm pretty sure this one part is not supposed to smoke......there, got it. Turns out that thingy with the fan on it goes the other way. Anyhow.

Wrong again, Zelig, I was communicating a single data point from the start, and wrong also, downtown, you can damn well sample anything you like. Tens of thousands of voters, or the district they're in, or the President they voted for. I chose to call it one district. One data point. Sample size of one. If you read some other meaning into my words, you're reading them wrong.

And everybody needs to stop reading wrong and start reading right in order to draw any lessons from NY-26. Starting small: every liberal in here who sees NY-26 as a Democratic turnaround in the making? You're wrong. You're reading tens of thousands of people into a nation of three hundred million. Wrong answer!! Both sample sizes--one district, and downtown's "tens of thousands" of voters--both of these are way too small a sample size to represent the entire nation. If you did that, you screwed up.

Moving on to my kind of scale. Big. Nationwide scale. Take a look at where the country has gone in the last ten years. From Bill Clinton to George Bush--to a Republican President and, for six of Bush's eight years, a Republican Congress. And from there to the exact opposite in 2008: a Democrat President and a Democrat Congress. Then, in 2010, the House flips back to Republican. Doesn't make a lot of sense, does it? If it doesn't make sense to you it's because you're reading the numbers wrong. You need to remember that all of these extremely dissonant election results were all caused by the same people. By the American Voter. A lot of the same people who elected George Bush (and then re-elected him) elected Obama. And those same people then turned on Obama in 2010. A lot of people cast American elections as "us vs. them", a war to destroy the Other Side, without even realizing that American voters are switching between those two sides all the time. Any politician who expects to make any headway with the American Voter needs to see this switching and figure out why Americans are doing it (and I already said--American Voters are doing it because they hate both parties).

So there it is. The two most important lessons we can draw from the NY-26 Special Election are these: the first lesson is that the politicians need to stop looking at districts and actually listen to the voters, and the second lesson is that everybody in CFC (except Arwon and BasketCase ) is really poor at analyzing elections.....

 

[Arwon]

downtown's "tens of thousands" of voters--both of these are way too small a sample size to represent the entire nation.

Why is that?

 

[Zelig]

One data point. Sample size of one.

You're the only person doing this, since if you know anything about statistics, this is obviously pointless.

 

[BasketCase]

Why is that?

One district is too small a sample size to represent the entire U.S. because the entire U.S. has 435 Congressional districts in it. Further: the United States is one-quarter liberal, one-quarter conservative, and half realists who either hate both parties or don't give a crap. Since an electoral district must always produce a result of "Democrat winner" or "Republican winner" (or in very rare cases an "Independent winner"), no one district can ever accurately represent the entire nation. Because no one district can ever elect a one-fourth Democrat one-fourth Republican half-Independent.

And tens of thousands of voters are too small a sample size to represent the entire U.S. because the entire U.S. has a hundred and fifty million voters in it. And guess what, there's a "further" here too! The second reason downtown's "tens of thousands" are too small a sample is because the voters in NY-26 who just elected a Democratic knucklehead are not themselves all Democrats. Some of them are, but the others are not--they're moderates, independents, or disgruntled Republicans who jumped ship. The recent election in NY-26 doesn't reflect any of this.

 

[Arwon]

Opinion polls manage with far fewer than tens of thousands. You can get extremely robust statistical estimates with a sample population of tens of thousands.

"The population voters in that district aren't representative of the voting population of the entire country at a general election" is a different argument, but one which has absolutely nothing to do with sample size. If a 50 000 person sample wasn't big enough to draw conclusions about a larger population, statistics simply wouldn't work.

 

[downtown]

Yeah, polling firms never sample tens of thousands of voters...that's simply too expensive. With a properly randomized sample, you can get quite accurate data with a few thousand. When I polled, our sample groups were often less than 1,000 (for a congressional district), and we were usually within 2 points.

Anyway, when we say "national lessons from NY-26", we don't mean that those results are applicable in every single district nationwide necessarily (or at least we shouldn't).

My congressional district, IL-4, couldn't be more different than NY-26. NY-26 is suburban and rural, blue collar, overwhelmingly white, old, somewhat socially conservative, and a place where Medicare is a significant political issue. My district is one of the most Democratic in the country (D+33 Cook Score), dominated by Hispanic voters, young, and WAY more concerned with immigration and and poverty reduction than medicare. Any political strategy used in NY-26 would be pretty irrelevant on the west side of Chicago, just like it likely would in St.George Utah, or Lake Charles, Louisiana.

The interesting thing would be to find districts that are demographically and politically similar to NY-26, to see if that political messaging can be generalized. Obviously, it isn't perfect, because there are too many variables, but there would be several races that are similar. If we believed that only Presidential Elections or nationwide congressional elections have predictive power, then we'd be basically saying that it is impossible to reasonably predict election results, since the electorate changes so much between cycles. We know that isn't true.