Trump Wins WI and PA and the Election
Updated: November 9 @ 9:11 am EST
Based on the polls, WI and PA should both have been Clinton states. For that matter MI also was in the Trump column, a state which has not been called as of this hour. In both states (WI and PA), Trump leads by about 1 point and in MI by about 0.3 points. These states should not have been close. I have chosen not to focus at all on why I think people vote the way they do, or discuss why people respond to polls the way they do. Simply put, the polls underestimated Trump support and the populations Democrats count on to pull out close elections, well they did not show up.
I am evaluating whether to begin a hub on the 2020 election. But at this point, I am wrapping up this years work and I thank my readers.
Election Day - Running Real-Time Information! [Final Update]
Updated: November 9, 2016
Note: All data are estimates from Slate.com/VoteCastr and reflect a model which assimilates existing real time turnout data, early voting and past election statistics. The estimates will change during the day as new information becomes available.
Points as to the success of this project:
- It was an experiment. The process was flawed during the day but I dd my best to keep readers alerted as to these projections. One of the problems that Slate/VoteCastr had was in separating early vote from same day vote. That would seem to be fundamental but was a problem. If they repeat the project in four years (maybe they give up, who knows?), they have learned a number of lessons.
- Slate/VoteCastr was always clear that the numbers were estimates and not projections. Of course, that did not stop anyone from looking at the numbers as projections. When I saw the early vote for FL (please see my early vote hub), I noted that the Republicans should have been pleased because the early vote for Democrats was down 2 points from 2012 (and Clinton lost in FL by 1.4 points). But then I saw these Slate/VoteCastr numbers and I figured the same day vote would save the day for the Democrats, and it did not. FL was way off the mark, as was OH, IA, PA and WI. So the Slate/VoteCastr people either need to adjust their model if they are going to do this again, or chalk up this effort to a Trump anomaly.
Most recent popular vote estimate based on data available at: 8:19 pm (EST) on Nov.8:
- FL: Clinton 4,959,569 Trump 4,644,007 (estimate reflects 117% of 2012 turnout)
- OH: Clinton 2,534,965 Trump 2,516,534 (estimate reflects 95% of 2012 turnout)
- IA: Clinton 659,498 Trump 645,935 (estimate reflects 85% of 2012 turnout)
- NV: Clinton 504,108 Trump 496,633 (estimate reflects 101% of 2012 turnout)
- NH: Clinton 311,833 Trump 289,125 (estimate reflects 86% of 2012 turnout)
- PA: Clinton 2,557,627 Trump 2,401,513 (estimate reflects 90% of 2012 turnout)
- WI: Clinton 1,366,876 Trump 1,193,332 (estimate reflects 85% of 2012 turnout)
9:58 am: Slate.com in conjunction with VoteCastr is introducing real-time data and information throughout election day. This is a first (they call it an experiment) and as such, there is no way to know exactly how data will be formatted, timing or any other feature. I will attempt to make sense of the information and summarize pertinent information right here on this page. You can check back late morning (after I have voted!) to see if I have been able to update any late breaking data or information. I expect to cover the race for President and Senate on this section of the page.
Analysis of Slate/VoteCastr Data
The following is my reaction to the above data [update #8], given the rest of the data from my hubs:
- You cannot tell what this means for the Senate. The state to watch is NH when the votes go live after polls close there. A Democratic win in both NH and PA would probably get the Democrats to 50 seats.
- Slate/VoteCastr had percents available for the four major candidates, but needed to take those charts down when they realized the charts did not include early voting data - a glitch on their part. If they fix this, I will provide percents and popular vote estimates. That is why I switched to popular vote. I also added the figure on the proportion of the current estimate to the 2012 total vote for Obama and Romney.
- Florida - these Presidential races in the recent past seem to always come down to Florida (Ohio is in 2nd place on this list). The current estimate has Clinton ahead by 277,000 votes and the total turnout estimate is almost the same as the total Obama+Romney vote when everything was counted in 2012, and we still ahve a few hours to go. So we appear to have a large turnout which includes the dramatic increase in early voting from 2012. Needless to say, if you are watching anything at this point, watch Florida!
- Iowa and Ohio currently favor Clinton but there is a long way to go. The party affiliation early voting data in IA was not as strong for Clinton as it had been for Obama in 2012. The early voting for Clinton in Ohio did not look strong although apparently they rallied at the end.
- The PA estimate favors Clinton, a state she is expected to win and a state that Trump does not need to win (if he wins FL and every other critical state)
- Similar to PA, WI looks good for Clinton. Trump made a challenge in Michigan. If the WI and MI data correlate, Trump may not win in the upper midwest.
- NH is tough to figure out. Clinton is ahead currently. Bernie Sanders and Elizabeth Warren have spent a lot of time there. Stay tuned.
- Nevada - what is going on? Well - cannot tell. Given the early vote, this should have been about a 5 point win for Clinton. The majority of people voted early and Clinton had a 7 point lead.
- Colorado is entirely ballot by mail - my probability chart says that Clinton should win and the last Slate/VoteCastr number was Clinton +5. Patience required!
Twelve Most Likely Outcomes!
Updated: November 8, 2016 [update #1]
Each day, I am updating the twelve most likely battleground states [FL, NC, PA, NH, OH, AZ, NV, CO, IA, and MI]. There are 1,024 possible outcomes when you determine the probability of each of these ten states being won by Clinton or Trump. Assuming all other states are won by the favorite, you start with Clinton at 224 and Trump at 180. Of course, the number of electoral votes required to secure victory is 270. Based on the very latest polling data, the twelve most likely outcomes follow:
- Clinton 323 Trump 215 [Clinton wins FL, NC, PA, NH, NV, CO, and MI] (p:3.70%)
- Clinton 308 Trump 230 [Clinton wins FL, PA, NH, NV, CO, and MI] (p:2.51%)
- Clinton 294 Trump 244 [Clinton wins NC, PA, NH, NV, CO, and MI] (p:2.32%)
- Clinton 341 Trump 197 [Clinton wins FL, NC, PA, NH, OH, NV, CO, and MI] (p:2.04%)
- Clinton 317 Trump 221 [Clinton wins FL, NC, PA, NH, CO, and MI] (p:1.71%)
- Clinton 279 Trump 259 [Clinton wins PA, NH, NV, CO, and MI] (p:1.58%)
- Clinton 329 Trump 209 [Clinton wins FL, NC, PA, NH, NV, CO, and MI] (p:1.51%)
- Clinton 312 Trump 226 [Clinton wins NC, PA, NH, OH, NV, CO and MI] (p:1.28%)
- Clinton 302 Trump 236 [Clinton wins FL, PA, NH, CO, and MI] (p:1.16%)
- Clinton 288 Trump 250 [Clinton wins NC, PA, NH, CO, and MI] (p:1.07%)
- Clinton 334 Trump 204 [Clinton wins FL, NC, PA, NH, AZ, NV, CO, and MI] (p:1.07%)
- Clinton 300 Trump 238 [Clinton wins NC, PA, NH, NV, CO, and MI] (p:0.95%)
Probability Trump Wins is 8.90%
Updated: November 8, 2016 [update #1]
Out of the 1.024 possible outcomes in the ten most competitive races, the most likely Clinton win scenario is 3.70%. The most likely Trump win scenario is 0.36%. The derived probability that Trump wins is now 8.90%.
Using the approach outlined above (1024 possible outcomes), what is the most likely combination of states which would result in a Trump victory and how likely is it? Please note this analysis assumes that Clinton wins ME cd2 and Trump wins NE cd2 (each cd has a single electoral vote). This analysis also assumes that Trump wins Utah. A McMullin win in UT could affect the final outcome.
The four most likely Trump win scenarios follow. What you quickly see is that the 3rd most likely Trump win scenario results in a tie. Presumably, since the election goes to the House (one vote per state), it is expected that a Republican controlled House would vote for Trump.
- Trump 279 Clinton 259 [Trump wins FL, NC, PA, OH, AZ, and IA] (p:0.36%)
- Trump 275 Clinton 263 [Trump wins FL, NC, OH, AZ, IA and MI] (p:0.32%)
- Trump 269 Clinton 269 [Trump wins FL, NC, NH, OH, AZ, NV, and IA] (p:0.29%)
- Trump 272 Clinton 266 [Trump wins FL, NC, NH, OH, AZ, CO, and IA] (p:0.10%)
With No Toss-up States: Clinton 323 Trump 215
Updated: November 8, 2016 [update #1]
Where does the presidential race stand today?
In terms of the battleground states, Clinton is ahead in: FL, NC, NV, PA, MI, VA, WI, MN, CO, NH, and the ME cd2. Trump leads in IA, OH, AZ and GA. In other competitive states, blue remains blue and red remains red. Clinton is ahead in WA, OR, NM, CT and ME; whereas Trump is ahead in SC, MS, MO, and TX. The states not listed are clearly predicted to vote as expected blue or red. So after all is said and done, the tally today is Clinton 273 and Trump 265.
What is the most likely path for Trump to get to 270 based on existing polling, if the dominoes fall for the trailing candidate (Trump) and against the leading candidate (Clinton)? Trump must win FL, NC, NV, OH, AZ, IA, and GA and then finally NH to get to 269 EV.
The situation is quite fluid and readers are encouraged to check back often for updates.
Technical notes: For purposes of this analysis, I use the average of weighted polls (sample size, recency and pollster historical accuracy are factored into the equation). I use a minimum of seven polls, but only polls from 2016. I use all polls from the last 30 days regardless of how many polls that is, assuming the number is equal to greater than seven. Every state, the Bangor Maine district (1 EV) and the Omaha Nebraska district (1 EV) have been polled.
With Toss-up States Included: Clinton 268, Trump 179, Toss-ups 91
Updated: November 7, 2016
Based on latest polling from a multiple of data sets, with tossups included, the race stands at Clinton 268, Trump 179, Toss-ups 91 [FL, NC, NH, AZ, IA, NV, OH, Omaha NE (1) and Bangor ME (1)].
How recent polls and the use of multiple data sets have affected the top line:
Please note that as of October 30, additional data sets have been added to the compilation of my statistics. I am now able to examine all 1,024 possible outcomes in the 10 most competitive races. If for example, WI becomes more competitive than MI, then I quickly substitute MI for WI in my data runs. The change to the bottom line above, in my analysis which includes toss-ups, is that NH is now declared a toss-up, and has been moved away from the Clinton column. In order for Clinton to win, she needs to hold the states where she leads, and win in one of the following: FL, NC, NH, AZ, IA, NV, OH or both ME cd2 and NE cd2. That would bring Clinton to at least 270.
Ohio has been a toss up state through the entire campaign. Ohio leaned to Clinton, then Trump, then Clinton and now Trump again. Using 37 polls completed in Ohio over the last 30 days prior to and including October 29, Trump leads by 0.11% (unweighted) and by 0.92% (weighted). These values are within the margin of error. OH is the most important state based on power ratings and has the smallest polling differential of any state in the country. With 18 electoral votes, this is THE state to watch.
In North Carolina, over the 39 NC polls conducted in the last 30 days, on and prior to October 29, 2016, Clinton leads by 2.21% (unweighted) and by 3.09% (weighted). As the variability in these polls is relatively small, both of the averages are outside the margin of error, and therefore we can declare NC now in the Clinton column, not a toss-up state.
In Florida, using the 47 polls from the most recent 30 day period, prior to and including October 29, 2016, Clinton leads Trump by 2.04% (unweighted) and by 2.59% (weighted). Given the large number of polls and the relative small amount of variability, both of these averages are outside the margin of error, Florida has moved from a toss-up to Clinton.
Pennsylvania has moved solidly into the Clinton column. Using 38 polls completed for PA over the 30 days prior to and including October 29, Clinton's margin is 5.03% (unweighted) and 5.77% (weighted). The Clinton unweighted and weighted leads are well both outside the margin of error. The reader can check out the capsule above to see the logical progression of states Trump would need to win if the dominoes start falling against Clinton. PA currently has the 7h smallest differential in polling and is rated the 7th most important state. If Trump does not win CO, he must win PA.
New Hampshire remains in the Clinton column. Based on the most recent 30 polls, prior to and including October 29, 2016, Clinton leads by 9.57% (unweighted) and 7.71% (weighted). Both averages are well outside the margin of error. NH has the 12th smallest polling difference and is assessed as the 12th most important state in this election.
Iowa is the state with the 2nd smallest polling differential and is now rated the 3rd most important state. Iowa has moved into the Clinton column. Using 16 polls conducted over the last 30 days, prior to and including October 25, Clinton now leads Trump by 1.31% (unweighted) and 0.14% (weighted). These values remain within the margin of error, and therefore IA remains a toss-up state, but IA is also a must win for Trump.
In Maine, the 2nd congressional district (cd 2) is clearly a toss-up. There have been only four polls conducted over the past 30 days including, prior to and including October 25, 2016. Clinton leads Trump by 1.00% (unweighted) and 0.60% (weighted). Both values are clearly within the margin of error. Without substantial additional polling, a projection for cd 2, and the associated one EV is clearly speculative. Clinton leads in ME overall (2 EV) and in cd 1 (! EV) by solid margins.
Recent Arizona polls have moved that race back into the toss-up column. Originally, I had AZ as a toss-up, then moved it to the Trump column, and now with the new polls, I have moved the race back into the toss-up column. Trump leads on average, over the last 30 days prior to and including October 25, 2016 (22 polls), by 0.86% (unweighted) and by 1.74% (weighted), both of which are within the margin of error. AZ has the 3rd smallest difference in polls between the two major candidates and the 4th highest power rating, making this state very important to watch. With its 11 electoral votes, AZ is a must win for Trump.
Nevada now shows the 6th tightest race in the country. Over 29 polls in the last 30 day period, prior to and including October 25, 2016, Clinton leads Trump by 1.66% using the unweighted approach. On a weighted basis, Clinton's lead stands at 2.12%. Both poll differentials are within the margin of error. Nevada is also a must win for Trump.
Based on the most recent 24 polls in Colorado, prior to and including October 25, 2016, Clinton's lead on an unweighted basis is 6.33%, and on a weighted basis is 5.80%. Based on poll variability and the algorithm used to develop the proprietary weighted analysis, the unweighted lead as well as the weighted average lead is outside the margin of error. This means that Colorado has moved into the Clinton column from being declared a toss-up state. The position of Colorado is quite important. Colorado holds the 8th smallest differential overall (after PA) and the 6th smallest difference for states in the Clinton column in state polling. Above you will note that Trump must win CO or PA to secure victory. Colorado is currently rated as 8th in the list of most important states, based on number of electoral votes and the distance a candidate must go to win the state.
Wisconsin polls remain relatively friendly to Clinton. Using all 24 polls in the last 30 days prior to an including October 25, 2016, Clinton leads by 7.14% (unweighted) and 6.11% (weighted). Both numbers are outside the margin of error which is why Wisconsin is not declared a toss-up state. Currently, WI has the 9th smallest polling differential, is the 10th most important state, and is the 2nd Trump alternative if he does not win PA or CO, using the domino approach above.
Michigan is regarded as a competitive state, although clearly in the Clinton column. Based on the most recent 23 polls over the 30 day period, prior to an including October 25, 2016, Clinton leads by 6.96% (unweighted) and 7.02% (weighted). MI has the 10th smallest polling differential and is judged to be the 9th most important state.
Minnesota receives relatively little attention from major polling firms. However, I was able to locate 15 polls over the last 30 days, prior to and including October 25. Clinton leads on an unweighted basis by 7.00% and on a weighted basis by 7.60%. These margins are well outside of the margin of error and continues to place Minnesota safely in the Clinton column.
Different organizations use different polls for their analyses. I attempt the best I can to aggregate the most recent polls for any state, regardless of the pollster. I do not use a single polling feed and this site is not automated. I look at every poll manually, time consuming yes, but I stop and think each time I analyze any piece of new information.
I have initiated a second process which I have incirporated into these polling hubs, that being weighting polls by sample size, recency of the poll and historical accuracy of the polling firm. I will cite both results of every analysis and make a special point if the results are different for any given state (unweighted and weighted). I am also changing the protocol for determining the number of polls I am using. if available, I will continue to use a minimum of seven polls. However, I will use as many polls as are available in the 30 days.
Please check back for updates as new polling data is released.
Clinton Leads in Battleground States by 4.71%
Updated: October 16, 2016
Only polls with ending dates on and after October 9, 2016 [63 polls] are included in this analysis. The polls from twelve battleground states (OH, FL, NC, NV, AZ, PA, CO, IA, MI, WI, MN and NH) are aggregated and the bottom line is featured in this analysis. The bottom line number will be updated every day until the November 8, 2016 election.
Clinton's weighted average lead across all twelve battle ground states is 4.06%. Please see below for updated poll weighted average differentials in every battleground state.
Poll Differentials in Competitive States
Updated: October 25, 2016
- Ohio: Trump leads by 0.07%
- Iowa: Clinton leads by 0.14%
- Arizona: Trump leads by 1.74%
- North Carolina: Clinton leads by 2.72%
- Florida: Clinton leads by 2.99%
- Nevada: Clinton leads by 3.12%
- Pennsylvania: Clinton leads by 5.66%
- Colorado: Clinton leads by 5.80%
- Wisconsin: Clinton leads by 6.72%
- Michigan: Clinton leads by 7.02%
- Minnesota: Clinton leads by 7.60%
- New Hampshire: Clinton leads by 7.78%
The data reflects weighted averages of polls conducted over the last 30 days. States are listed with the smallest poll differential at the top. Only competitive states are listed.
State Power Ratings
Updated: October 25, 2016
- OH 100.00%
- FL 83.77%
- IA 82.99%
- AZ 76.30%
- NC 72.43%
- NV 55.78%
- PA 51.06%
- CO 36.34%
- MI 34.44%
- WI 29.42%
- MN 21.39%
- NH 9.13%
Note: The computation of state power ratings combines how close the race is with the number of electoral votes associated with the state. The states at the top of the list are more winnable by either candidate and more important overall. Each state below the most important state (currently OH) is rated as to significance as a proportion of the top state. Example - PA is 51.06% as important as OH.
Cross State Comparisons
Updated: October 5, 2016
The question I am investigating is whether polling data in one state is related to polling data in another state. If Candidate A performs relatively well in one state, can you then make a prediction as to how well that candidate will perform in another state? There are three categories of outcomes with correlations: Candidate A could poll relatively high or low in both states A and B (that is a positive correlation - the closer the correlation is to 1.00, then we have stronger confidence in making a prediction; or Candidate A can poll relatively highly in state A and relatively poorly in state B - that is referred to as an inverse correlation and the closer that inverse correlation is to -1.00, the greater the confidence we will have in making a prediction; or there appears to be little or no relationship in polling date across states for candidate A - in this case the correlation will approach 0.00. That is a brief explanation of how correlations work. One important caveat - you can never assume causality -we are looking at relationships only.
Florida vs. Pennsylvania
As of October 5, 2016, over the last 30 days, there are 16 polls in common between these two states, which means there are 32 data points (16 polling numbers for Clinton and 16 polling numbers for Trump). The value of the correlation is very high at 0.85. Taking the statistics one step further, this means that FL poll numbers account for 73% of the variation in polling data for PA and visa-versa. How a candidate does in FL is related to how that same candidate will do in PA and visa-versa.