I got that giddy feeling seeing all those Biden 10% leads in recent national polls and in that slew of NY Times/Siena College state polls. Did you?
5-Second Takeaway
Polls should make us confident to work hard campaigning, but not complacent. Those NY Times polls all had a margin of error around 4.2%, so those 10% leads are less than my safe threshold of 3xMOE. Only good for Likely Ds, not Safe Ds.
Biden 350, Tossup 62, Trump 126.
The 270 Threshold continues to sit in the Likely Ds somewhere between FL, WI and NH.
Hot contests are TX, GA, IA and OH.
Swing States (yes, these states are traditionally red)
GA Fox released a June 20-23 3% MOE poll today. (You may be surprised that Fox is rightly rated A- by 538!) Last poll was June 12-13. Both show Biden leading by 2%. That’s Leans D (2/3 MOE) and just barely Leans D (½ MOE) The May 21-27 -4% poll has timed out for me. As of the Fox poll release this afternoon, Georgia has moved to Leans D!!! (2 Senate races, Tossups!)
TX Forget Trump and their governor’s mishandling of the pandemic. Texas is trying to move back into the pink as a PPP poll shows Trump with a 2% lead, just over half the MOE. They showed it tied just 15 days earlier, June 3-4, so that’s a trend against Biden. But Fox came out today with Biden +1!
538 has a very nice average over time graph. Its time decay generates a trend that lags actual changes in polls. They showed the race tightening even before today. Trump’s lead down to 0.2%.
270ToWin only uses the latest poll from each pollster and only from the latest month, then does a simple average. So they throw out PPP’s tie and average Fox-s +1%, Quinnipiac’s May 28-Jun 1 -1% and PPP’s Jun 18-19 -2% to get a 0.67% Trump lead.
My method looks at all 3 in two different ways. First, to avoid blending poll MOEs, I categorize each one based on its own MOE. Fox is +1 vs. 3 < MOE/2, so it’s a Tossup. Quinnipiac is -1 vs. 2.9 < MOE/2, Tossup. The first PPP is 0%, obviously a Tossup and the latest is -2 / 3.3 > MOE/2 => Leans R. So that’s Tossup, Tossup, Tossup, Leans R. Secondly, I do a simple average to get a Trump lead of 2/3% with a median MOE of 3.3%. < MOE/2. Tossup it is! (Senate race Likely R)
IA June polls were -1%, 0, -1%. Trump lead < MOE/2. Tossup (Senate race Leans D!)
OH The numbers were there on June 2, when a Fox poll put Biden at +2% with MOE of 3.5%, for Leans D. But now Quinnipiac shows Biden +1% with MOE of 2.9%. Biden lead, but Tossup.
Baby pink
MT The only poll since February put Biden 5 points behind on May 5. Sitting Gov. Bullock's entry into the Senate race flipped that race and made this one competitive. Leans R (Senate race Leans D!)
UT Trump’s lead sunk to 3% on May 26. Maybe Mormons’ sense of decency will overcome politics.
AR The only presidential poll out of Arkansas put Trump at a 2% lead on June 9-10. Leans R.
Baby Blue
NC was Leaning D in April and May, and it is again now. (Senate race Tossup.)
AZ I am ignoring the very small polls from Change Research, rated C- by 538. That gives Biden a 4.75 point lead. 538 figures 4.8 pts and 270 figures 4.0 pts. Any of those is good for a bit more than 1 x MOE. Leans D (Senate race Safe D!)
NV Just 3 polls this year, the last one at the end of April, was good for Biden 4%, just above the MOE.
A word on my analysis methods…
I take all polls I can find from 270ToWin or 538 for the latest month polled, omitting any pollsters 538 rates C or below (like Change Research and Gravis) or that have a noticeable bias (like Gravis or Rasmussen). To avoid a simple average without working a weighted average, I categorize each poll to get a predominant category. I check that against the category of a simple average of the polls.
Stale states will stay in the same category until new polling comes in.
I categorize leads by this arbitrary set of rules:
My definition of a Tossup is a lead of +/- ½ MOE. That means the leader has less than a 60% chance of truly leading. See Fig. 1 below.
I define Leaner as between ½ and 3/2 of the MOE, so the reported leader has between 59.9% and 77% chances of actually leading. See Fig. 2 below.
My Likely category is between 3/2 and 3 * MOE, so the leader has between 77% and 93% chance of being ahead.
My Safe category is only for leads at least thrice the MOE, so the leader has a greater than 93% chance of being ahead.
I am not doing any predicting, just blending any and all polls within the last month (or last month taken — if the last poll was in April and it’s June, I’ll take the month of April) rated B/C or better by 538.
This is intended to give a snapshot of the horse race. Nothing more.
Figure 1. Lead = ½ the margin of error. Leader polls 48%, Trailer 46%, MOE 4%. Assume normal distribution and Trailer increases exactly match Leader decreases (undecided is constant). Probability of Leader really leading is the probability that his true percentage is 47 or higher (hence, the probability that the Trailer’s true chance is 47 or lower). From the graph, Prob = 59.87%
Figure 2. Lead = 1.5 x the margin of error. Leader polls 48%, Trailer polls 42%, MOE 4%. Assume normal distribution and Trailer increases exactly match Leader decreases (undecided is constant). Probability of Leader really leading is the probability that his true percentage is 45 or higher (hence, the probability that the Trailer’s true chance is 45 or lower). From the graph, Prob = 77.34%
Try out this and other scenarios at onlinestatbook.com/…
