Long thread: One of these is not like the others. The generic ballot average or trend has taken a dip recently, but it is largely driven by a dramatic change in one poll, the Ipsos/Reuters poll. 1/ 
Ipsos/Reuters has been in line with other generic ballot polls and not unusually volatile over the past year. But since May 1, the Ipsos/Reuters results have turned sharply towards the GOP. 2/
Here are the Ipsos/Reuters results using their 5 day rolling sample since they began asking the generic ballot May 23, 2017. These, and all results in this post, are for registered voters. (h/t to Reuters for making this available) 3/
Obviously something shifted on or about May 1 to coincide with a sudden shift from a relatively stable mean +7.0 prior to May 1 to a mean +4.2 after that date, with considerable volatility. 4/
What could it be? Did whatever changed also affect Trump job approval? Yes. Yes it did. Like most polls, Ipsos/Reuters shows Trump's gradual climb since early December with a consistent -20 to -10 net prior to May 1. Then a spike. 5/
Did rising Trump approval account for the shift in generic ballot? No. No it did not. The far more likely explanation accounts for both these shifts. The partisan composition of the Ipsos/Reuters sample changed sharply on May 1. 6/
The partisan balance in Ipsos/Reuters polling shifted suddenly from a mean of +6.3 to a mean of -0.7 after May 1. That 7 point swing in net partisanship of the sample surely drives both generic ballot and Trump approval shifts. 7/
There have been short term shifts in partisan balance in the Ipsos/Reuters sample, but all have been brief with a quick return to the long-term mean. This chart shows all Ipsos/Reuters rolling 5 day polls since Jan 2017. 8/
Short term noise in party ID is normal and while it makes individual samples bounce around it is quickly gone as new daily samples come in and the data quickly returns to around its long term mean. 8a/
This shift in partisan balance is unlikely to be due to a sudden shift in the actual partisanship of registered voters. Why there is this sharp change is not clear. Ipsos has not changed it's methodology, so far as I know. 9/
Or possibly this reflects real, unexpected, change. There are three weekly generic ballot polls, all from online pollsters: YouGov, Morning Consult and Ipsos. Real change should appear in all 3 but YouGov & Morning Consult don't shift. 10/
A side note is the limited number of live phone polls. Since May 1, there have been just 2 live phone (completed 5/1 and 5/5) and 12 online polls, plus two IVR polls from Rasmussen. Our most frequent generic ballot data is from online. 11/
Online polling is becoming a large proportion of polling on the 2018 election. This example shows it is more important than ever to monitor mode and pollster effects, not to dismiss outliers but to diagnose them. /12
Ipsos is a top-notch polling organization & they've developed a good non-probability online methodology. This shift is a puzzle. Online non-probability polls are still challenging. This is a chance to lean more about how they work/fin
In the last 4 days, Ipsos/Reuters party ID has moved up to a Dem +4.1 margin, and generic ballot has moved up to Dem +7.2.
