Why the Harris +5 Marist national poll is useless

| Political Ref | October 17, 2024 | Permalink

With a Poll Bias Score of 70 toward Dems, this national Marist poll from 10/8-10/10 released on 10/16 reflects definite bias and I needed to adjust the topline for my average. As always, the pollster topline goes into the pollster average. The poll included far too many Dems, a defect that fatally pollutes all other ID categories.

When a category contains too many Democrats, the gender, race, age, region or level of education loses relevance because an inordinate percentage of each category will lean Dem no matter what. So if you flood a poll with Dems, who cares if it's a man, non-college grad or a Hispanic person? I adjusted the poll based on how the pollster determined each party will vote, which was D's 96-4 Harris, I's 54-44 Trump and R's 94-6 Trump. See calculations here. Based on party vote, Harris gets 49.2 to Trump 50.3.

You will hear the argument that one shouldn't cherry-pick cross-tabs, and that's true. But two categories, party ID and education ID have the potential to ruin a poll if they stray too far from the likely turnout, so that no amount of balancing of biased categories matters. To those Harris supporters reading this, if you read a poll showing a five-point Trump lead based on an R+7, you would laugh and potentially get upset, and justifiably so.

My averages include as many polls as possible. If a pollster's sample substantially diverges from the most likely turnout, I adjust the sample and simply apply the pollster's vote ratios to each voter ID category, e.g. age ID, gender ID, etc. If I includ polls that depart from the 2020 exit polls as severely as Marist did on party ID, my average would offer little usefulness. I aggregate a massive sample using public polls with similiar samples, either natively or after I adjust the sample, achieving a result that begins to approach the campaign's internal polling accuracy.

Marist's results actually support a Trump lead but their topline shows a +5 Harris lead. Seems like an overreaction to the pro-Trump data they were collecting. Dems need to realize that this sort of thing does not help you in the end. It may help Harris raise $1 billion, but at this point some percentage of hard to motivate Harris voters see that and think, well she's got it locked up so I don't need to vote. If you don't think that's true, you don't know Democratic voters.

It also serves to anger observant Trump supporters. Dems goal for GOP voters is to make them feel sullen with falsely positive Dem results, but there's a fine line. When you release obvious outlier polls designed to throw off aggregates it angers Trump voters and then they start memeing harder than ever. You want a sullen Trump voter, not an enraged one.

So in the end, a poll result based just on party ID is not as reliable as one with a good cross-section of voter ID categories all balanced to match the most likely turnout. But if that's what Marist wants to provide, that's what I will take. Do not pretend that this Harris +5 result means anything but the Democratic academic establishment that Marist occupies is freaking out. It could not be clearer.