The USC “Daybreak” poll is the new Rasmussen, and something very fishy is going on which doesn’t seem to jibe with the rest of the polling world.
If you’re not aware of Trump’s new favorite poll, their approach is this. USC selects a nearly fixed panel of 3000 (they actually slowly build over time), not unlike the RAND panel from the last election, and they repoll ~1/7 of their group each day. Consequently, the numbers represent something like a 1 week rolling average of respondent sentiment.
In principle, this should be great for looking at trends. While the panel may be biased one way or another (and boy it sure is!), changes in the numbers from day to day or week to week represent actual individuals changing their minds, rather than statistical sampling fluctuations.
But there are some very weird features. For one, the poll starts with an extremely R-leaning bias. On July 10, the first day of their data, they have C-3, when the Huffpo pollster had it at C+4 — that’s a 7 point Republican bias. As you’ll see in the snapshot above, according to USC this morning, it’s C-7.3, which points to a nationally tied race once we “unskew the polls.”
Note: I do not want us to act like the republicans from 2012 in decrying skewed polls! But it’s at least worth looking into the results in greater depth.
Individual cross-tabs also seem to have odd results. For concreteness, compare the most recent USC results with the latest Economist/YouGov poll (which I chose because of ease of cross-tabs; 538 only gives YouGov a B rating, and claims a D +1.6 bias).
- USC has Trump ahead in all age ranges, including C-5 for 18-35 year olds. Compare this to YouGov, which puts C+20 for under 30’s (not a perfect comparison).
-
USC has Trump getting 37% of the Hispanic vote, to 50% for Hillary, (C+13), While YouGov has it 64-25 (C+39). Also, compare this to Obama in 2012, who bested Romney 71-27 among hispanics (O+44). It is really hard to imagine Donald (“build a wall”) Trump doing significantly better among Hispanics than Romney.
Two important groups, and a 25 point difference between the two polls.
I’m even more confused by the trendlines overall. Like it or not (and I do not) most polling outlets suggest that over the last 2-3 weeks, Trump has gained by about 4-5 points on Hillary. That’s certainly what the Pollster yields. But what’s strange about the USC panel is the timing.
The RNC began on the 18th, and for the next two days (during which panelists were reacting to what they saw), Hillary actually gained in the poll. For two days after that, she gave back those gains, leaving the poll roughly the same at the end of the RNC as at the beginning. But thereafter, it’s been a steady stream upwards for Trump, and downwards for Hillary, where we now stand at C-7.3. This, despite a terrible few cycles for Trump (Russia!), general party unity at the DNC (much better than expected, no thanks to the California delegation, but much thanks to Bernie being magnanimous) a bunch of really well received speeches (Michelle!), and on and on. It’s very hard to imagine that during the first three days of the DNC voters were so appalled by what they saw that ~2% of them decided to switch from Hillary to Trump.
USC saw an 8 point shift in the last week! Let’s compare that to some other polling agencies that took tracking polls both this week and last (shift in results): Reuters (C-1.5), YouGov (C-1), SurveyMonkey (Unch). We also saw Rasmussen jump C+6, but that was over a 2 week period. And none of these have any potential DNC bounce timed in yet.
So I ask again, WTF is going on here?
