People don’t do this. Cross tabs have giant margins of error AND aren’t weighted within the crosstab. “The crosstabs look weird” just isn’t evidence on the accuracy of the toplines either way
In the NYT/Siena poll of Wisconsin likely voters, head to head: - 18-29 year olds are voting for TRUMP by 18 points - 30-44yo are voting for HARRIS by 21 points - the 44-59yo are voting for TRUMP by 9 points - the 60+yo are voting for HARRIS by 15 points www.nytimes.com/interactive/...
Results of New York Times/Siena College poll of 680 likely voters in Wisconsin conducted from Sept. 21 to 26, 2024.
Congrats are in order for whoever got 18-29 year-olds to answer a call from an unknown number
Abramowitz used to say whatever the sample error for the survey double that and you might get to the cross tabs
Can you say more about this? Specifically: are you saying that even if the crosstabs like that 18-29 one look wacky, that isn't inherently a reason to question the poll?
I would love for the industry to build norms around common definitions for crosstabs (like, where do you make cuts on age distribution to create buckets) so we can better average these together. But that’s a separate issue
Is there an explainer about this anywhere like an animation, youtube, whatever? I feel like I want something short and substantive above and beyond "If a public poll were that bad, you'd see partisan pollsters rightfully criticize it."
Did the Times disclose if or how many poll respondents were at cell phones? Were these suspect ‘18 to 29-year-olds’ all at land lines?