Likely voters or registered voters?

Readers may have noticed that PolitickerNJ.com and the Asbury Park Press have different numbers in reports in ledes on the Monmouth University/Gannett New Jersey U.S. Senate poll story. PolitickerNJ.com went with the 45%-37% among likely voters, while the APP led with the tally for registered voters: 39%-28%. Both are valid numbers, and this website would be interested in hearing from our readers on their preference: should we use likely voters or registered voters when reporting poll numbers?

Likely voters or registered voters?