Monmouth Poll: Clinton +4 in Ohio

[Alcon]

The thing is Seriously?, 2012 exit polls and actual polls of likely/registered voters showed there to be more Democrats than Republican in Ohio. Why would Monmouth find there to be more Republicans?

They got a crap underlying sample here. That's my point. There was a lot of massaging to make this thing even workable. I am well aware that the exit poll numbers were in the D+7 or so range. They massaged this poll to D+4. The raw sample was R+4.

Did they massage this poll so that it ended up D+4, or did they do other demographic weights that happened to bring the sample to D+4?  There is a huge, huge methodological difference between those two things.  Obviously, neither is desirable, but with polling response rates what they are these days, demographic weighting is probably becoming more and more pronounced in polls.  That may make it more desirable than declining to re-weigh.

[Alcon]

The thing is Seriously?, 2012 exit polls and actual polls of likely/registered voters showed there to be more Democrats than Republican in Ohio. Why would Monmouth find there to be more Republicans?

They got a crap underlying sample here. That's my point. There was a lot of massaging to make this thing even workable. I am well aware that the exit poll numbers were in the D+7 or so range. They massaged this poll to D+4. The raw sample was R+4.

Did they massage this poll so that it ended up D+4, or did they do other demographic weights that happened to bring the sample to D+4?  There is a huge, huge methodological difference between those two things.  Obviously, neither is desirable, but with polling response rates what they are these days, demographic weighting is probably becoming more and more pronounced in polls.  That may make it more desirable than declining to re-weigh.

Demographic weighting is industry-standard, not undesirable. Even in the days of 25% response rates, (good) pollsters weighted because of differential turnout among RV subsamples, and unequal selection probabilities within households.

To be clear, I didn't mean to indicate weighting is a bad practice -- it's a necessary one.  I just meant polls would be even better if the phenomena that necessitate weighting didn't exist.  It's undesirable that poll respondents are so unrepresentative (and it's getting worse!). While weighting is the best solution to fixing that problem, it would be better if the problem didn't exist.

For instance, we can weigh up the sample of 18-to-24 year old white females, but that assumes that we're getting a representative sample of 18-to-24 year old white females, and whatever causes a low proportion of those to respond isn't also causing an unrepresentative sample of those to respond.  Weighting is the right thing to do in that case, but it can't fix every underlying problem.

edit: A more obvious example - weighting up the Hispanic respondent % will get you a misleading result if you don't give a poll in Spanish, because English-only interviews will result in an unrepresentative sample of Hispanics, and re-weighting on race alone won't fix that.

[Alcon]

Seriously?, I think your post raises good questions, although I tend to defer to competent pollsters on matters like this.  I just have one thing to add:

One would think "voter reg history," would be taken into account in the LV screen.

Not necessarily.  Most likely voter screens entail asking questions about self-reported voting history, self-reported intention to vote, self-reported interest in the election, and sometimes, screening questions that determine if the voter has a plan to vote, e.g., knows when and where to do it.

The thing is, these self-reports are more likely to be accurate if they correspond to past behavior than if they don't.  At the end of the day, a 0/4 voter could express high intention to vote, high interest, and even high voting history ('cause people are big ol' liars).  A 4/4 voter could give the same information.  Which voter would really be likelier to vote?  Obviously the 4/4 voter, not the 0/4 voter.  So it makes sense to weigh down the 0/4 voter's response, just because we know that 0/4 voters who give the same answers as 4/4 voters are still less likely to vote.

Also keep in mind that meeting the LV screen does not mean all respondents score equally highly on it.  If the screen determines one voter has a 99% chance to vote, and another 75%, it makes sense to allow both through the LV screen.  But it also makes sense to weigh down the 75% person's response to reflect the fact that their "predicted vote" for their preferred candidate is 0.75 votes, as opposed to 0.99.

Make sense?