Oct 10, 2012

Statistics and manipulation of the ignoramuses (contd.)

[Previously in the series]

Why do politicians and news organizations poll?

The simple answer to this query is to gain some insight into how the population might be considering a particular issue. For reasons of costs and logistics, you cannot seek a ballot of the entire set, so you conduct a poll. You pick a sample size, you determine a margin of error, you craft the questions, you strive to eliminate bias, and arrive at a measure.

In other words, statistical science works, bitches!

For instance, Gallup reports that Americans are not in favor of any particular set of values over the prevalent opposing view, that is to favor some particular set of values called "traditional values."

By a margin of 52 to 44. The US population is 310+ million, and while the adult population is a little smaller (250+ million), getting all those many adults to vote is obviously still too expensive and a logistical nightmare.

So statisticians at Gallup chose a random sample of about 1,000 people, and they estimated the margin of error to be some +/- 4%.

All it means is that Gallup estimates that the population of the US today is likely to reject traditional values and prefer no specific set of value by a margin of 52 to 44. Of course, some 4% are unsure, or provide non-responsive answers. [Update: This is incorrect, see first comment below]

Polling is of course also used to evaluate the state of the Presidential race.  Mitt Romney was trailing significantly till his magical debate performance, which has not only stemmed the tide but has massively reversed it. To the extent that he is now leading in a few national polls where he was trailing before.

Naturally this has excited the rightwingnuts beyond belief, when just a fortnight ago they were screaming conspiracies in polling.

But now a pollster appeared on FOX "News" so that he can further feed the audience what they want: continuing positive news for Romney.

The article Pollster: Romney’s Already Won Florida, Virginia, North Carolina cites this:
PALEOLOGOS: We’ve already painted those red, we’re not polling any of those states again. We’re focusing on the remaining states. 
Paleologos said the movement towards Romney in the three states was “overwhelming.” Obama was in particular trouble in Florida, where even before last week’s presidential debate he was only up by 46 percent to 43 percent over Romney.
So it'd stand to reason that the pollster (Suffolk) had conducted polls to establish this, right?

No, it does not appear so. The last poll Suffolk conducted in Florida was conducted on 10/2/2012 (one day before the debate), and it shows Obama with a slight lead: 46 percent to 43 percent, with 7 percent undecided.

So what made Paleologos conclude that Obama was now so far behind that he's not even going to poll there? It must have been the preceding poll and the trend, right? Well, given that he polled just one other time since May and it was not in Florida, but in Virginia, not very likely.

Answer: Nothing that he or his organization polled.
Real answer: Likely budget problems, so guess as appropriate. Polls cost, and he's out of money for now.

This is precisely the bullshit that objective, alert media anchors would point out. If they were even pretending to be objective. It is FOX "News" with Bill O'Reilly fer cryin' out loud! This was never about delivering news, but propaganda.

[Next in the series]


rappoccio said...

One small nit:

All it means is that Gallup estimates that the population of the US today is likely to reject traditional values and prefer no specific set of value by a margin of 52 to 44. Of course, some 4% are unsure, or provide non-responsive answers.

The 4% uncertainty includes the statistical uncertainty on the sample size. If you queried everyone, then this would be "zero" because it's simply the result. However since the size of the query is only 1000 or so, then, assuming that the 1000 is unbiased relative to the whole, then the fluctuations will be governed by a binomial distribution. Thus, if you repeated the query many times (but still over 1000 people, different people each time), you'd expect a variation, just from statistics, of around 4%.

PS for the details, the 95% confidence-level binomial error on a fraction of 52% in a sample of 1000 people is

error = 2 * sqrt(0.52 * (1-0.52) / 1000)
= 0.04

Staid Winnow said...


The point I was making was that the Suffolk pollster was asserting that he did not need to poll Florida anymore because Romney has an insurmountable lead.

How does he know this?

Was there a series of polls?


Was there even ONE poll he conducted that showed Romney way ahead in Florida?


How many polls had he conducted?


What did that poll show?

That Obama was ahead.

So he had NO polls he conducted that showed Romeny ahead, one (only one since May) that showed Obama was ahead, and his conclusion?

Romney is so far ahead that he will not be polling anymore.

rappoccio said...

Yes, absolutely, I was just nitpicking ;)