This poll is atrocious for Perdue, loads of bad signs for him:
He's already consolidated the Republican base whereas "Democratic opponent" still has 15-20% of the Democratic base to fall in line.
A combined 47% of independents say they "need to know opponent" or are "not sure" - essentially both meaning undecided.
Moderates are overwhelmingly against him.
This poll is not atrocious for him.
-this poll is clearly oversampling democrats because no, in the reality the GA electorate is not D+1, far from it. Even in 2018 the GA electorate was still R+4
-this poll has Biden winning by 4, which is almost impossible considering how hard it is for democrats to crack their statewide ceiling of 48%
No, this poll is atrocious for Perdue. Paying attention to toplines only is not the proper way to read a poll - toplines are almost always the result of weighting which is entirely based on the pollsters' assumptions. This has a much smaller impact on crosstabs. Any good pollster advising a campaign will spend 10% of their time on the polling conference call talking about toplines and 90% talking about the crosstabs.
The crosstabs here are perilous for Perdue. He is getting demolished among moderates and is drastically under where a GA Republican needs to be with independents a year out, especially for an incumbent (in 2014, our numbers had Deal ~15% higher among independents than Perdue is right now at the same point, and we considered that to be a sign that the race would be competitive). He is in incredibly bad shape and has a lot of work to do.
Also, "oversample" is a strategic polling technique that means to collect an additional group of responses from a demographic of interest (i.e. racial minorities, people under 24, people in a certain media market) that is then weighted appropriately into the full sample in order to get a more accurate read on that demographic and ergo the entire electorate. What you are referring to is "overrepresentation" - a poll giving more weight to a certain demographic than it warrants - which will almost never actually effect the accuracy of a poll (unless it is extremely drastic) because of the margin of error and the central limit theorem.
I don't understand your whole point, but it's totally stupid for a pollster to over-represent some groups (democrats and minorities in this case) just because they could be ''demographic of interest'' ; first of all why do such groups are interesting from a polling perspective ? I mean black voters and liberal democrats will vote for the democratic candidate no matter what, a demographic of interest would be center-right suburban voters who are probably the only kind of swing voters in GA ; but blacks and democrats are not by definition swing voters. Besides, if they want to look at the voting patterns and political interests of democrats and black voters they should simply do a separate poll ; but by definition when you oversample some democratic constituencies in you general election poll you will end up with a result which is biased toward democrats and that's what happened with this poll
You are confusing oversampling and overrepresentation again. Let's say I'm polling for a campaign in Georgia. By default for a statewide survey, I'd want a base sample of 800. African Americans typically respond to polls in lower numbers, so it's likely our collected sample will not have enough to represent their share of the vote accurately. We can weight them up in the final sample, but weighting up is generally dangerous since you are giving outsized weight to a small subsample that has a larger margin of error. To combat this, we will collect an additional 100 responses from only African Americans (an oversample). We then weight this 100
into the 800 base. The extra AA responses are weighted down to their appropriate share of the electorate, and the 900 total responses are weighted down to 800 to avoid giving extra weight to non-AA responses. This gives an accurate overall sample, as well as a much more accurate subsample of AA voters.
Overrepresentation would be if we just included the additional 100 without correcting any of the weights, or if we just didn't collect any additional responses and had the wrong percentage of Black respondents. Even then, over/underrepresentation does not necessarily ruin a poll (unless it's a massive error) due to the central limit theorem.
