New York Times overhauls Times-Siena polling to better reflect voters

The New York Times said it is overhauling the Times/Siena poll with a new weighting method aimed at making survey results more deeply representative of the population. The change, disclosed June 29, uses a technique that asks, in effect, how many people in the population are like a given survey respondent, a sharper way of deciding how much each answer should count.

The shift is already visible in the June 2026 Times/New York Times/Portland Press Herald/Siena poll in Maine. That survey interviewed 608 registered voters, reached 93 percent of them on a cellphone, and weighted the results to better match the likely electorate by accounting for vote propensity, past voting behavior, and characteristics such as gender, age and education. In plain terms, the poll did not just count responses equally; it adjusted for who was most likely to show up and who was less likely to be overrepresented in the sample.

That Maine survey was the first in the Portland Press Herald’s collaboration with the Times/Siena project, and readers were given access to the full tables of results. The design shows how methodological choices can change the picture of public opinion before a headline is written. A poll that reaches voters mostly by cellphone and then rebalances the sample by voting history and demographics can produce a different read on the electorate than one that relies on looser assumptions about who is likely to vote.

Siena Research Institute remains the polling engine behind the project, and its June output shows the operation is active across more than one state. A Siena release on New York State politics, dated June 25, surveyed 1,120 registered voters from June 17 through June 23 and used dual-frame landline and cell phone interviewing, text-to-web outreach and a proprietary online panel. Those mixed modes, paired with the new Times methodology, point to a polling operation that is trying to become more precise about both who gets reached and how each respondent is weighted.