California Polling

There's a new USC Dornsife/LA Times poll out and I'm not going to link it. Why? Because it's such an embarrassing poll for USC that they should really quit polling. It's an entirely online poll. There are good reasons to do online polling but polls that were entirely online produced some really funky results in 2016. They aren't as accurate as polls that include phone calls.

The poll was conducted between April 18 and May 18. I've never heard of doing over a period of a month because the data is inconsistent. The people who answered in the early part of the poll could've made up their mind or even voted by the later part of the poll. Their answers are meaningless. That's why polls are conducted over 3-4 days.

This poll has roughly 40% undecideds in both the gubernatorial and senate race. That's really strange so close to the election and inconsistent with other recent polls. A poll with that many undecideds doesn't tell us how people will vote. The 2014 primary had 4.5 million votes. So far over 500,000 ballots have been returned. So 10-12% of the voters have not only made up their minds but their vote is set. At this point a pollster should have two samples, people who have voted and people who will vote. That'll give them an accurate idea where the election is.

California political data expert Paul Mitchell on why pollsters are so wrong on California polling. California has a ton of data on who votes. Yet pollsters ignore it. Mitchell shows the share of voters by partisanship, geography, and ethnicity. Latino turnout in primaries has hovered around 12%. Yet the pollsters are showing on average a 24% Latino electorate. They look at what share of registered voters Latinos are and decide that's who is going to vote.

California has party registration. So we know the share of the electorate each party had in the past and we can get a good idea of how that fits in this year by looking at how the VBM returns compare to previous years. Right now it's 44%D/33%R. That's slightly more Democratic than 2014, 44%D/36%R, but less than 2016, 49%D/34%R. Mitchell points out that Republican average 31% of the voters in the public polls but have been 38% of the voters in the last three gubernatorial primaries. Republican share could certainly be lower this year due to Republicans re-registering as NPP but that's a lot lower.

Getting party ID correct is vital in top two. The biggest primary question is whether two Democrats make top two in the senate and gubernatorial races. A poll with 30% Republicans may miss that.