Poll Analysis

TAKING A BREAK UNTIL AFTER THE FIRST DEBATE

Isn’t this just like the 2012 “unskewers” that predicted Mitt Romney would win?

Another thing to consider is that Obama overperformed the final polls in 2012, which is evidence that Obama had some of the “monster voters” that I discuss in this article. This time, if anyone gets “monster voters” who pollsters tend to underpredict, it’s going to be Trump – i.e. Trump is more like Obama, Hillary is more like Romney.

Instead of multiple posts, this will be a page with poll analysis, updated periodically.

“Unskewing” formula:

Party	2012 ID	2012 Turnout	ID/Turnout	2016 ID	2016 projected turnout
Democrat	35	38	1.086	30.0	32.58
Republican	30	32	1.067	27.7	29.56
Independent	33	29	0.879	40.6	35.69

So I took the 2012 party ID / 2012 turnout, and then multiplied the turnout for each group by 2016 party ID. This gives the projected turnout for each group in 2016.

Then when I see a poll, I look for party ID breakdowns, and weight how Republicans, Democrats and Independents respond to pollsters. I then weight the responses of each group based on the above projected turnout.

Party ID is from Gallup.

Analysis of all August Polls of Huffpo Pollster that had party ID breakdowns, and Rasmussen, LA Times and Dornsife.

Rasmussen and USC/Dornsife are included because they have transparent methodologies: Rasmussen strictly weights by party ID, USC/Dornsife tracks a population which are a representative sample of voters in the 2012 election (i.e. percent that voted for Obama vs. Romney). The LA Times tracking poll describes their methodology thusly:

Results are weighted to match demographic characteristics, such as race and gender, from the U.S. Census Current Population Survey, and are aligned to the 2012 presidential election outcome using self-reported votes in that election.

Italicized numbers are adjustments made when only the weighting data of a poll is given, but not the full party ID breakdowns (i.e. how many Republicans plan to vote for Trump, Hillary, etc.), and based on previous polls, I simply give Trump 0.9 points for every point weighting of democrats above D+3, as I believe the general election will be D+3. For state polls I use the gallup numbers instead.

National Polls

Poll	Unadjusted Numbers	Unadjusted Gap	Adjusted Numbers	Adjusted Gap
Ipsos/Reuters 9/8-9/12	39-39	Tie	36.73-35.03	T+1.7
Rasmussen 9/15	42-40	T+2	None	T+2
USC/Dornsife 9/15	47.4-41.0	T+6.4	None	T+6.4
CBS / Yougov 9/4-9/6	40-38	H+2	38.49-37.97	H+0.52
IBD/TIPP 8/26-9/1	39-39	Tie	38.74-38.16	T+0.58
CNN/ORC 9/1-9/4	45-43	T+2	43.79-40.21	T+3.58
Franklin Pierce 9/6	43.7-41.4	H+2.3	42.06-41.74	T+0.32
GWU/Battleground 9/7	42-40	H+2	37.92-36.83	T+1.09
ABC/Wapo 9/10	41-46	H+5	45.31-40.26	H+5.04
NBC/SM 9/5-9/11	42-40	H+2	X	X
UPI/CVOTER	49-46	T+3	X	X
Morning Consult 9/6-9/8	43-44	H+1	43.76-40.59	T+3.17
Quinnipiac 9/8-9/13	41-39	H+2	41.09-38.78	H+2.31
CBS/NYT 9/9-9/13	42-42	Tie	39.69-39.29	T+0.4
PPD 9/12	44.7-41.4	T+3.3	43.77-34.15	T+9.62
FOX 9/11-9/14	41-40	T+1	37.98-39.23	H+1.25
Average		T+0.09		T+1.41

Using only polls with an end date in September

Effect of my adjustments: Trump gains 1.32 points

Battleground polls have a quick and dirty weighting for party ID. I presume 36% Independents, and the remaining 64% split by Democrats and Republicans. I then use Gallup’s party ID/Lean by state, and add 2/3 of the difference between Democrat and Republican to whoever has an ID advantage.

Wisconsin (D+3)

Poll	Unadjusted Numbers	Unadjusted Gap	Adjusted Numbers	Adjusted Gap
Marquette 8/25-8/28	41-38	H+3	N/A	N/A
Monmouth 8/27-8/30	38-43	H+5	39.74-42.54	H+2.8

Michigan (D+5)

Poll	Unadjusted Numbers	Unadjusted Gap	Adjusted Numbers	Adjusted Gap
Detroit News 7/30-8/1	41-32	H+9	38.15-34.85	H+3.33
FOX 2 / Mitchell 8/9-8/10	49-39	H+10	47.5-46.58	H+9.08
Suffolk 8/22-8/25	43.6-37.2	H+6.4	41.1-39.7	H+1.4
Emerson 8/29	39.5-44.9	H+5.4	N/A	H+2.7
FOX/Mitchell 9/6-9/7	45-39	H+6	34.82-34.52	T+0.3
Detroit Free Press 9/15	38-35	H+3	X	X

Ohio (0)

Poll	Unadjusted Numbers	Unadjusted Gap	Adjusted Numbers	Adjusted Gap
Monmouth 8/23	43-39	H+4	41.72-40.12	H+1.6
CBS/Yougov 8/21	46-40	H+6	43.68-41.69	H+2.0
OnMessage 8/13-8/17	43-42	H+1	47.56-42.28	T+5.28
Emerson 8/29	43.4-42.5	T+0.9	N/A	T+3.6
Quinnipiac 9/8	46-45	T+1	45.88-45.8	T+0.08
CBS/Yougov 9/7-9/9	46-39	H+7	41-42.64	H+1.64
Bloomberg 9/14	43-38	T+5	X	X
CNN/ORC 9/7-9/12	44-40	T+4	43.56-40.4	T+3.16
Suffolk 9/15	41.6-38.6	T+3	X	X

Pennsylvania (D+5)

Poll	Unadjusted Numbers	Unadjusted Gap	Adjusted Numbers	Adjusted Gap
Emerson 8/29	42.7-45.9	H+3.2	N/A	H+1.0
Monmouth 8/26-8/29	48-40	H+8	44.13-42.12	H+2.01
Franklin and Marshall 8/29	47-40	H+7	40.11-38.32	T+1.79
CBS/Yougov 8/30-9/2	45-37	H+8	42.29-37.49	H+4.8
Quinnipiac 9/8	48-43	H+5	51.04-42.03	H+9.01

Virginia (D+1)

Poll	Unadjusted Numbers	Unadjusted Gap	Adjusted Numbers	Adjusted Gap
CBS/YouGov 8/2-8/5	37-49	H+12	41.12-43.4	H+2.28
Washington Post 8/11-8/14	51-43	H+8	50.91-41.58	H+9.33
Hampton 8/24/8/28	43-41	H+2	N/A	N/A
Emerson 8/30-9/1	43.5-43	H+0.5	N/A	T+2.2
PPP 9/9-9/11	39-45	H+6	40.57-41.93	H+1.36
U of Mary 9/15	40-37	H+3	X	X

Nevada (0)

Poll	Unadjusted Numbers	Unadjusted Gap	Adjusted Numbers	Adjusted Gap
CBS/YouGov 8/2-8/5	43-41	H+2	47.52-35.81	T+11.71
Suffolk 8/18	44-42	H+2	40.24-45.58	T+5.34
NBC/WSJ/Marist 9/6-9/8	42-41	T+1	43.44-39.2	T+4.24

Florida (D+3)

Poll	Unadjusted Numbers	Unadjusted Gap	Adjusted Numbers	Adjusted Gap
Monmouth 8/12-8/15	48-39	H+9	50.2-37.36	H+12.84
CBS/Yougov 8/10-8/12	45-40	H+5	44.04-40.56	H+3.48
FAU 8/23	43-41	T+2	N/A	T+5.6
Fox 13 8/10	45-44	H+1	N/A	Tie
Mason-Dixon 8/22-8/24	44-42	H+2	43.4-40.2	T+3.2
FL Chamber of Commerce 8/24	44-41	T+3	N/A	T+4.5
PPP 9/7	44-43	T+1	42.87-38.31	T+4.56
Quinnipiac 9/8	47-47	Tie	47.21-44.44	H+2.77
JMC Analytics 9/7 – 9/8	46-42	T+4	47.01-39.83	T+7.18
CBS/Yougov 9/7-9/9	44-42	H+2	43.5-40.57	T+2.93

New Hampshire (R+3)

Poll	Unadjusted Numbers	Unadjusted Gap	Adjusted Numbers	Adjusted Gap
CBS/Yougov 8/10-8/12	45-36	H+9	44.4 – 38.98	H+5.42
WBUR 8/20-8/28	43-32	H+11	35.64-32.74	H+2.9
Emerson 9/3-9/5	42-37	H+5	N/A	H+1.5
NBC/WSJ/Marist 9/6-9/8	39-37	H+2	42.45-42.27	H+0.18

North Carolina (D+2)

Poll	Unadjusted Numbers	Unadjusted Gap	Adjusted Numbers	Adjusted Gap
Quinnipiac 9/8	43-47	H+4	46.36-46.42	T+0.04
Suffolk 9/5-9/7	44-41	T+3	47.18-38.35	T+8.83
CBS/YouGov 8/30-9/2	42-46	H+4	43.49-43.48	H+0.09
Civitas 9/11-9/12	42-42	Tie	43.95-39.23	T+4.72

Facebook Comments

Daniel Balint

“So I took the 2012 party ID / 2012 turnout, and then multiplied the turnout for each group by 2016 party ID. This gives the projected turnout for each group in 2016.”
Let me make sure I understand correctly; Did you multiply the 2012 party ID / 2012 turnout ratio by the 2016 party ID? If so, since you multiplied by a number greater than 1, how can your projected 2016 turnout rate for Rs be lower than the 2016 identification rate for Rs?
- Ryan Faulk
  
  It’s not, republicans are 27.7%, not 30.0%. I put in the cells wrong.
BadgerWA

Informative thread. I understand poll unskewing failed in 2012. What were the adjustment factors then, and how far were they off?

I think a critical factor in 2012 was many Republicans and conservative leaning independents stayed home, because Romney was such a weak candidate. Obama was visible relieved, somewhat shocked that he won.

If Trump is to win, a critical factor will be a greater turnout among Republicans and conservative independents, and a lesser turnout among Democrats and progressive leaning independents. I am confident that turnout for Hillary will be lower, as I detect no ardent support for her except from the media. I’m sure Trump will motivate many who stayed home in 2012 to come out and gote for him, and he’ll get the vote of some progressives, but the NeverTrumpers are a problematic faction.