My dude is getting bad at the one thing he was actually good at.

My dude is talking about the very large sample Harvard Youth poll that specifically polled young voters aged 18-29 lol. In other words, not crosstabs, but the topline polling of young voters in a very high quality poll.

I agree crosstab-diving is generally unwise. Namely because subgroups in a poll have smaller sample sizes with larger margin of error than the topline poll and often lack the weighting of the topline.

Think of when you see a poll of 1,000 people that only polled 50 black people. That would have a margin of error of 14% for that subgroup, even though the poll had a MOE of only 3%. Effectively useless results in a horse race poll as MOE in a horse race poll applies to each individual candidate's support, so would be doubled for the difference. And that's just for sampling error.

Know who does crosstab-diving all the time? And did it recently regarding young voter subgroups? Nate Silver on his blog lol.