Polling in Virginia, while sparse, has generally shown a narrow lead for a Democratic-aligned amendment to the state’s Constitution that would temporarily put redistricting back in the hands of the General Assembly.

Early voting in Virginia showed high turnout, similar to what was seen in last year’s governor’s race. But redistricting is not a strictly partisan issue, and poll results have ranged from the amendment being ahead by eight percentage points to it losing by eight points, with a relatively small percentage of voters undecided.

Polls ahead of California’s similar redistricting measure last November showed more broad support, but there were still large shares of undecided voters until about two weeks before the election, leaving some Democrats anxious about whether turnout among their voters would be low, and if those who did vote would refuse to back the measure on principle.

Both ballot measures are intended to combat President Trump’s efforts to generate Republican districts through redistricting in other states. In Virginia, the referendum asks voters to revisit a 2020 decision in which they overwhelmingly voted to create an independent redistricting commission, stripping the General Assembly of its mapmaking power.

Now those same voters are being asked to temporarily reverse that decision in an attempt to eliminate all but one safe Republican district in the state.

As California’s recent election demonstrated, accurately polling referendums can be difficult. Beyond undecided voters, another central issue is the wording of these survey questions. Ballot language is often intentionally complicated in referendums, leaving pollsters in a tricky position: They can use the actual ballot language and risk confusing respondents, or they can devise their own cleaner wording, which has the potential to skew results. Two of the six polls in Virginia used the actual ballot language.