So you are just pulling numbers out of thin air. You don’t know the total population of Latinos or the population of people who crossed a border without going though a port of entry or over stayed their visas.
I’m using direct observation and interaction with people in my city. Bayesian inference based upon a sizable sample size across many years? If you have a source of immigration status of Spanish radio listeners by US metropolitan area, it would be an interesting number. Someone has those numbers using more quantifiable methods than my sampling technique.