Tuesday, August 9, 2016

Taking Your Polls With a Grain of Salt

One of the harsher criticisms delivered by a New Testament author is found in the Letter to the Hebrews. (If you aren't a "believer," I hope you'll tolerate this brief observation because it leads somewhere important.)
The writer criticizes Hebrew believers that aught, based on the passage of time and experiences of life, to be eating a diet suitable to adults. The writer uses the phrase "strong meat" to describe the diet that the writer expected those Hebrew believers to be of sufficient maturity to consume. (For the vegan reader, that's going to take you sideways if you get locked up on it … maybe think of the writer as having endorsed Brussel sprouts.) The writer's point is that, as adults, from the time our milk teeth drop out and our permanent teeth drop in, we are capable of eating more than breast milk from our mothers.
The writer's criticism focused on those Hebrews that should be, spiritually speaking, on an adult diet of spiritual understanding, but who were, instead, still feasting at the teat, surviving on the barest spiritual nutrition.
“Time to grow up,” I think, was the author's point.
Elections seem to bring out the teat sucker in folks.
I have in mind the consumption of news about presidential polls.
So, in the span of a few short days, we have been invited, courtesy of traditional news sources, to believe that we have swung from a Trump bump to an electoral Armageddon in which the Democratic Beast rises to the throne of America. Is that so obviously the case? If you take the Pablum of predigested summations, you may think it is so.
I think not. And, if you are a thinking person, you may think not too.
If you are a person that isn't content to avoid macerating facts, if you don’t simply accept the spoon fed agenda of the lap dance media, you will have looked through their filters and studied and understood the implications of what is, and what is not, being reported, about polling.
It's time to grow up.
If you have read this far, then I know you are capable of reading and understanding basic written English; to my thinking, that also means that you are capable of getting beyond the Pablum of media reports about presidential polling, that you can, you should actually take up the source information and screen it for its meaning.
Because it seems to be the place where folks shop for polling data, let's use RealClearPolitics and its polling summary page as an example. I've reorganized and reproduced their summary here:

 As shown, the largest sample polling was that of the Los Angeles Times poll. That poll, of over 2000 participants, showed the narrowest margin between Clinton and Trump, a single point of preference separating the two.
Moreover, that poll, based on its sample size had a self-reported negligible margin of error. Finally, among the polls summarized, that poll sought the views of LIKELY VOTERS, and was one of only TWO polls in the RealClearPolitics summary that sough the views of so-called LIKELY VOTERS.
The other poll of LIKELY VOTERS was completed a week prior to the LA Times Poll. It included a smaller polling sample, produced a reportable margin of error, and, while it showed Clinton ahead of Trump, the spread between them was nearly completely covered by that margin of error.
The remaining EIGHT POLLS were all of small sample sizes compared to the Los Angeles Times poll.
The remaining EIGHT POLLS were of a different demographic group. Where the LA Times and Reuters/IPSOS polls sought the views of LIKELY VOTERS, the remaining polls all sought the views of REGISTERED VOTERS.
Give some thought to what this could be telling you.
As a nation, a substantial minority of us, in many cases nearly 50% of those who are not only old enough to vote but are also eligible to vote do not vote. The most recent presidential election in which both the number of registered voters and the number of actual voters is certain, 2004, showed a turnout of about 60% of registered voters. You can check my math here and here.
Of course, there are many uncertainties in every election that can cause variations in the turnout and the outcome. Yet, if the search is for the most likely accurate forecast of how the election actually would turn out, you have to ask yourself, would the more accurate picture be gained by polling those who are registered to vote? Or, would the more accurate picture be gained by polling those who are, based on prior behavior and declared intentions, actually LIKELY TO VOTE?
The wildly ranging numbers in the REGISTERED VOTER POLLS, it seems to me, are exactly what to expect in a large scale group amongst whom nearly 50% are unlikely to actually vote in an election, even though they have bothered to register to vote.
Here’s another curiosity for a “meat-eating” poll reader:
If a poll of REGISTERED VOTERS showed one kind of trend or condition, but a contemporaneous poll of LIKELY VOTERS produced a contrary trend or condition, what meaning, if any, might/could/should we draw from a preference for polls of REGISTERED VOTERS or of LIKELY VOTERS?
Suppose that polling REGISTERED VOTERS produced reliably more accurate results, results that actually matched election outcomes. Or suppose, to the contrary, that polling LIKELY VOTERS did so. In either event, for what possible reason would a news agency or its polling partners choose to gauge the less reliable group?
Think about it.
Suppose you had to make your living betting on the outcomes of NFL games.
If you did so, which of the following would be more important to you: knowing what would be the actual starting line-up for Sunday's game, or simply knowing the undifferentiated list of all 53 eligible players on the team's roster?
To be sure, when injuries require that they do so, coaches will go into the bench; and you will want to know, in turn, about the depth of the bench for various positions, but in the first instance, how likely are you to pick one team over another, or even more so, pick a game based on point spread, if you are told only that there will be 11 unidentified players on one team playing against 11 unidentified players on the other team. The risk of losing on your bets would be enough to make you seek legitimate employment!
If the coach puts a passing quarterback on the field, that is some indication of the game to be played.
Here, the generic "registered voter" is placed in comparison with the more certainly known "likely voter."
If you think it odd for me to focus on the differences to be found in REGISTERED VOTER polls and LIKELY VOTER polls, I’d recommend reading some coverage of the differences. You can read about this topic, in the context of previous elections, here, here, here, here, and here.
And if a media organization likes the odds spread when polling REGISTERED rather than LIKELY voters, and then actually produces only polling of REGISTERED voters, that should give you that one more insight:
Yes, the media does have a preferred candidate, and they don't like the trends among likely voters, so they disguise them by polling a fairly irrelevant group instead.