Just saying that this poll is composed of 10% or so more Democrats than Republicans, but this is still trouble for Don
First of all, there are more Democrats than Republicans in this country at present, so this is reasonable. Would you contend that there should also be as many Libertarians sampled as Democrats and Republicans? Second....
Begin: soapbox
Not to pick on you particularly, because we see this kind of complaint frequently in this forum and elsewhere, but this is a completely bogus complaint. Yes, it's reasonable to reweight samples for demographics such as sex, race, and age, but these are all inherent or "fixed" attributes of a voter. (That is, it's reasonable if your turnout assumptions by demographic are good -- which is a whole separate question.)
The problem with complaints about more Democrats or more Republicans, in a sample is this:
Party ID is not a fixed attribute.Although some people have a strong allegiance to one party or the other, many people
don't. (And even some who do loyally identify with one party may still vote for the other, as shown by other discussions on this forum.) Party ID is, in a sense, nothing more than a personal opinion; it's not an unchangeable characteristic such as race. Most important for this discussion: a voter's party ID can fluctuate in the short, medium, and long terms.
For the short term, just take a look at the NBC/Survey Monkey polls before, between, and after the conventions. Between the conventions -- during Trump's bounce -- the party ID of their sampling became significantly more Republican than it was before the RNC. But after the DNC -- when Clinton bounced back -- party ID shifted back to near the pre-convention proportion. The samples shifted because the ID of the general population shifted. This is because many voters' party ID follows their preference for President,
not the other way around. Such fluid voters don't have a strong party ID going in, but their thinking (consciously or not) runs along the lines of "I support Trump, therefore if someone asks whether I'm a Democrat or a Republican, I must be a Republican."
For the medium term, look at the posters on this forum who have changed avatars on this forum during this campaign. If you took a census of the party ID of all members a year ago, the proportion identifying with each party would look different than it does now. Therefore, a representative sample would also have shifted.
For the long term, many people's political leanings shift over the course of their lives, due to age, income changes, life experiences, etc. I know personal anecdotal evidence is unscientific, but I'm sure I'm not the only one with a similar story. I was raised as a Rockefeller Republican (a species now unfortunately extinct), and in my life I've identified at various times as a Republican, a Democrat, a Libertarian, and an Independent. I've voted for all of these in Presidential elections; the Independent was John Anderson -- in fact I was an Anderson volunteer. (Yeah, I'm getting up there.
)
Creating a voter model and weighting for various demographics essentially boils down to trying to create a function with several input variables, such as age, sex, race, location, income, etc., and the output being a valid prediction of the vote. I submit that party ID is one of the
outputs of this function, not one of its inputs.
Because of this, questioning the party makeup of a sample is highly dubious. Doing so leads to the kinds of "unskewing" that failed notoriously in 2012. In general, most samples at the time showed fewer Republicans and more independents than in 2008. In hindsight, this was during the rise of the Tea Party movement. Many of the TPers were frustrated with the Republican establishment, and chose to no longer identify with the party. However, they still voted very much like Republicans. As such, this made the Independent group more favorable to Republican candidates. This would not be a problem, if people then didn't attempt to manipulate the numbers. But the unskewers applied the results of sampling based on 2012 party ID to a turnout model based on 2008 party ID -- when it was clear (at least in hindsight) there had been a significant chunk of Republican-voting people whose party ID had shifted from Republican to Independent. Their voting intention was the only important thing, not their party ID. This is why the unskewing attempts were so off base; this technique is completely invalid.
Disclaimer: I'm not a professional pollster, and this is just my personal opinion based on several decades of observation of American politics. On the other hand (we moderates always have to balance!), I have a graduate degree in systems analysis and work as an engineer for a Fortune 50 technology company; I know a fair bit about sampling and statistics.
End: soapbox
