The Daybreak tracking poll differs from traditional polls in two major respects. Rather than questioning a different group of respondents for each poll, the survey relies on a panel, currently consisting of about 3,000 people recruited at random to represent U.S. households.
The panel is part of a larger Understanding America Study conducted by USC's Dornsife Center for Economic and Social Research. The election survey is being done in partnership with The Times and USC's Jesse M. Unruh Institute of Politics.
Because of the panel design, "we have the same people every time, so changes in the poll are really people changing their minds," rather than the result of variations in who answers a particular survey, said Arie Kapteyn, the director of the USC Dornsife center, who pioneered the approach for the 2012 election while at Rand Corp.
The panel design typically shows less volatility than traditional polls. Four years ago, it proved more accurate than most other surveys in forecasting the election result, although "maybe that was beginner's luck," Kapteyn said.
The other major difference is that the poll, using a 1-to-100 scale, asks respondents to say what the chance is that they will vote as well as the chance that they will cast a ballot for Clinton, for Trump or for another candidate. The results are weighted based on those probabilities, so that a voter who is 100% sure of his or her choice has more impact on the forecast than one who is 60% sure.
That approach is one way to resolve "one of the biggest problems that polls have deciding who is going to vote," Kapteyn said.