The proper method is to keep dialing randomly until you get a large enough group of Latino voters to complete the survey.
Then I can compare the answers of say White and Latino voters, and retain the ability to weight them back to the entire population, when looking at the population in total.
I'm not saying that a poll can't be skewed, even on purpose. But, over and under sampling is a valid procedure.
That being said, the pollster is paid by someone. I'd never put my name on a skewed poll, but I can't vouch for all of us.