Reply #20: They didn't actually release the Hispanic subgroup. The author of the article you are talking about

[BzaDem]

computed an approximation of the Hispanic subgroup given other data.

The reason they don't release it is because subgroups have much smaller sample sizes, and the smaller the sample size goes, the more error you get. Most polls have some subgroups that are wildly divergent from reality, but their top line ends up being accurate.

(For example, if you were to take a 1000 LV sample, and then look at the subgroup of people with blue eyes and red hair, your margin of error on that subgroup will will be in high double digits, even though the margin of error on the whole sample is only around 3%.)

I think their likely voter methodology is underestimating turnout. But the small-sample Hispanic result you are talking about isn't the main reason.