Princeton Election Consortium

Innovations in democracy since 2004

Nov 03: Biden 342 EV (D+5.3% from toss-up), Senate 53 D, 47 R (D+3.9%), House control D+4.6%
Moneyball states: President AZ NE-2 NV, Senate MT ME AK, Legislatures KS TX NC

Any requests?

November 2nd, 2008, 10:30pm by Sam Wang

I’m taking requests for analysis to do between now and Tuesday morning.

Possibilities include the following:

A cheat sheet to get you through Election Night. Presidential, Senate, and House races to watch, and early warning signs of what the evening will yield.

Analysis of inefficiencies and errors in electronic markets. I would also name contracts that are under- or over-valued. (Even if I post, if you send me proof that you have my book or have donated to one of my favorite races, I’ll send you recommendations by late Monday.)

Final Presidential, Senate, and House predictions. Actually, you’ll get this one no matter what. I’ll post them late Monday night, when I enter them into various contests.

A chart for applying biases and “corrections” to polls. This is in case you believe in the Bradley effect or want to assign undecided voters, and want to see how it affects the prediction.

More snark about other aggregators. I’m sure many of you want this one, right? (crickets)

Is there anything else?

Tags: 2008 Election

62 Comments so far ↓

  • Mkpowers

    I’m interested in seeing a full “The Power Of Your Vote” chart. I live in California so it’s interesting to see how few Newmexicovotes or Coloradovotes my 1 Californiavote is worth.

    That, and I want to see you and Nate Silver in a cage match.

  • JeffH

    I’d like to see a chart that lets us calculate probable outcomes based on popular vote adjustments. For example, I believe that Obama is actually 11% ahead nationally based on the major outlets’ final polls. I’d like to see how that would affect the states, and which would flip. Others may like to see the map if Obama is only ahead 2% or something. Thanks Sam!

  • Geza Gyuk


    Fantastic site. Between your site and I’m in polling heaven.

    I’d appreciated your post “An under-the-radar…”. Could you generate the same plots but for 2004? The comparison for the full time line would be interesting.

    Something else I’d find interesting:

    a) a density plot of the frequency of n-percentage point changes in the meta-margin over m days. Essentially a measure of the historical frequency of shifts of a given size over a given time period.

    b) Same thing but for the median electoral college estimation.

    Thanks again for a great site.


  • bill meadow

    I vote for #1 — the cheat sheet;

    I’ve been checking your site every 8 hours for months; it’s terrific;


  • Eliot Greenwald

    I would like a cheat sheet and final predictions for president, senate and house.

  • BCC

    I love the snark about other aggregators! I also enjoy calling you on it.

    I will say that this site beats 538 hands down for the signal-noise ratio in the comments. Being slightly less discovered makes for a higher-quality commenting community.

  • DFS

    It seems you have tended to poo-poo the effects of election fraud (primarily because of the margin you see), but there continue to be stories on numerous outlets, such as BradBlog and Huffington, as well as Political Carnival, documenting vote flipping, voter intimidation, inadequate resources dedicated to handle the volume of expected voters, random registration issues, etc

    I’d like to see an analysis of why you think these issues wont be determinative and any debunking that can be done of the allegations.

  • GAW

    As an intrader with some free margin right now, I’d love to see some analysis on inefficiencies in the electronic markets.

  • Jay

    I’d like to see an analysis of presidential election outcomes that includes data for third party candidates. There’s a lot of talk right now of undecided voters breaking for McCain. But almost no analysis of how third party votes effect the margins. On, in their chart that includes third party candidates, it seems like they have consisetnely hurt McCain, since the beginning of the general election. So does this offset any extra margin McCain might get from undecided voters? In any case, it seems like any real world analysis of possible outcomes needs to include third party votes, since some small percentage of the population (but perhaps significant in close states) will vote for third party candidates.

  • George

    I like to see the % chance of Obama winning all of the 2004 Kerry states PLUS Iowa; Colorado; and New Mexico.

    IOW, the chance of just getting 273 EVs.

  • Paul

    Yes, final predictions, of course! A line in the sand makes it all twice as fun.

    Might be tricky, but a chart to answer this question: If we see unexpected results from various states, do they fall along a uniform national deviation from the expected, in a meta-margin-style shift? Or are they peculiar to certain states?

    Post election: an analysis of the quality of various polling organizations. Are all polls created equal? Re-run the meta-analysis with different organizations included and excluded. Would removing any particular organization actually have *improved* the meta-analysis’s prediction? Or is a kitchen sink mix of all polls truly the best?

  • George

    I like to see your take on which East Coast states to watch on Election nite like New Hampshire; Virginia; North Carolina; and Florida.

    IOW, how early could a TV viewer tell which one will win: Obama or McCain?

  • Arnold Evans

    Is there enough information to measure the value of one phone call, one knock on a door and one dollar contributed in terms of votes?

    I read at 538 somewhere that assuming one leaves New York and goes to Pennsylvania to canvass, every 12 contacts nets a vote.

    How does that translate to making calls or donating money?

    More improbable that enough data is available:

    How much of the difference between Obama’s polling or his results and those of Kerry would you calculate came from the difference in money? How much from other factors such as the economy or a better volunteer force?

  • The Liberal Crab

    I’d like to see an analysis that regresses each poll to the same party id weighting and overall % of total. It’s something I wanted to play with, but since I am not one of you guru’s and have no access to subscriptions, I haven’t been able to play that game. It would be interesting to see how all the polls, corrected for weighting, balance to each other.

  • bks

    If Obama wins by 9% of the popular vote and loses in the electoral college, where is the safest place to hide?


  • Charles Yeganian

    Cheat sheet, definitely. What else am I supposed to pass out at the election party to counter my wife’s stupid Oscar ballots?

  • Eddie

    A simple one: What is your own, personal, subjective estimate of McCain’s chances?

    Is it the same as the model (.00…0007%)? Or 20%? Or 10%?…

  • Michael McIntyre

    I’d love a cheat sheet that:
    (1) provides a guide to key house races throughout the night, and
    (2) projects national totals based on deviance of early reporting states from your final Monday predictions

  • Eddie

    Also: If you could run your model Monte Carlo style just to more clearly see the difference between yours and 538’s?

    And if you could run it with uncertainty for undecideds?

    Thanks for all the hard work you’ve shared!– it’s been very enjoyable and interesting and impressive!

    I read a cool psychology/neuroscience article this month, and am now interested in reading your book.

  • Jeremy

    I would also like to vote for option #1 – the cheat sheet.


  • DavidP

    I would like to see an election night cheat sheet so that I can put more meaning behind the numbers that roll in.

    I am also interested in some tool that would help detect and estimate a systemic bias in the polls (e.g. Bradley/cellphone).

    By the way, I followed you every day back in 2004 and I have checked your meta-margin several times every day since August of this year. Wonderful work! I especially appreciate the absence of “corrections” to your raw numbers. 538/RCP are very interesting, but I put more faith in your blind analysis.

  • SM

    At this point, I’m more curious about key House and Senate races than the presidency — it seems like a clear result on the exec branch. I’m curious to see exactly which seats you expect the Ds to pick up to get to 59 in the Senate. (The chart is great, but I want to know which of the tight elections you think will get the Ds to 59.) Thanks!

  • Owen

    I agree that there is a lot of pessimism about the certainty of this race coming from the left. With that in mind, let’s suppose the results on election day are very different from all of the poll aggregating sites (all of which show obama with more than 300 EV right now). Have you considered what some possible causes of this potential outcome would be? Do you think there is anything relatively significant that poll aggregation misses?

  • Fred

    Perfect. Exactly what I was hoping to find. Here’s what I’m after:

    It’s great to know which Senate races to keep an eye on, since that is my main point of interest this election. What I can’t seem to nail down is WHERE one can go to hear such determinations (or even solid guesses) as they happen. A person can watch just about any news channel Tuesday evening and probably hear from each of them, within seconds, who won Florida or Ohio. The Senate, which (ask Clinton) matters perhaps more? Not so much.

    I’d also love to hear some sort of analysis or guess as to what percentage of “independents” elected to register as such in hopes of avoiding the scenario where GOP-influenced folks in charge of such processes may have chosen to “lose” the registration. I fall into that category, as a case in point.

  • ron

    Have you figured out (figuring poll closing) which state will put him over? of course you would need an equation telling the computer when a state was likely to go after its polls close. Also what will the total number of voters be, it would be the best measure of interest as you could measure total increase of voters against historical increases.

  • David Palmer

    I don’t post here much (most of you are MUCH better at statistics than I am) but I check this site repeatedly during the day and enjoy it very much.

    I vote for the cheat sheets.

    But snarky comments (particularly about one site in particular) would also be VERY welcome.

    And thanks for all the hard work!

    David in Albuquerque

  • Mark M.

    I too vote for cheat sheets: if this turns out to be a 3-sigma result, where will it show up first? N. Carolina?

    By the way in case I forget in the (hoped for) euphoria….Thank You!

  • Matt A.

    I’d love the cheat sheet.

  • Rorschach

    Something I’d like to ask about: is there a way to track increases in straight-ticket voting this cycle?

    The apparent dissatisfaction with the GOP in general makes me wonder how large the coattail effect might be on Congressional races. If Obama wins by a large margin, I find myself wondering if the chances of a filibuster-proof Dem Senate majority are greater than anticipated.

    I’m no number-cruncher but I’m curious. Thanks.

  • Observer

    I’m with those who would like to see analyis of the effect of third-party candidates in this election. (I’m guessing not much, given O’s overall lead, but you will know far, far better than I.)

  • FrankS

    I am interested in more “what if” scenarios concerning poll assumption problems. Showing different distributions for different skews would give a sense for how stable the current electoral vote is to such underlying assumptions.

    Another topic is one I doubt you have any data for …

    There have been lots of assumptions mentioned in the news and blogs (e.g. party distribution, likely voter models, cell phone only households, age distributions, etc.) and it leads me to wonder if anyone has studied the effect of the caller ID label on poll results … does getting a call from “University of X” result in a different outcome (different answering population) than getting calls from “unknown” or “Rassmussen”)?

  • James

    It would be really interesting (but perhaps a lot of work) to see trend lines for each state normalized wrt the national trends. E.g., (using totally fabricated numbers) in August Michigan was 4.5% more pro-McCain than the national average but by October it was 0.6% more pro-Obama.

    The interest would come in matching those trends with per-state campaign expenditures in time and money.

    The effect could be to isolate the effect of advertising (including campaign stops) from more global trends such as the mortgage crisis.

  • Rachel Findley

    I just made up my own amateur cheat sheet, so I could start guessing whether it’s an historic landslide or a mere victory. I even looked up the poll closing times, so as to know when to scrutinize the results from Georgia (-3), North Carolina (+1) and Virginia (+4) to see if there’s a surprise. Then I remembered I’ll be out working to get out the vote until 8:30 PM CA time, and the national election will presumably be over by then.

    I’m sure anything you do will be superior to my efforts, and will be appreciated by people who are watching the results.

    I noticed that you and Silver at fivethirtyeight have different win probabilities for many states where the results are not yet certain. His are generally lower, including, spectacularly, Florida, which you give a 98% win probability and he says is 50-50 (or did I copy something wrong?).

    I know what your probabilities are based on: just the polls. But I’m curious why you think so many people, not just Nate Silver, think there’s something that doesn’t show up in the polls that is pushing toward a more even race.

    I’m wondering if there is a post-election test that would indicate which set of probabilities was closer to the real probabilities. You would want to specify the test before the results roll in.

    Mind you, I’m glad that many Obama supporters are running scared. I’ve been putting all my energy into opposition to Proposition 8, which would write a ban on same-sex marriage into the CA constitiution.

    If all were as (over)confident as I am, we could just manage to lose the election due to failure to get out the vote.

  • Sherry

    Yes, Sam, can you please talk about your take on Florida for obama? You seem to be the only site predicting that and the polls seem close. I would be very interested in your thoughts about this.

  • Gretchen

    I would like to see some analysis of the relationhsip between complaints of problems at poll sites in 2004 and this year’s election, and/or adding in the effect of getting Obama election observers and lawyers.
    And thank you for your excellent site!

  • Martin

    When you post your predictions, I’d really appreciate the values for the each of the most likely outcomes on the EV outcomes chart.

    Thanks for another great election cycle — I hope you’ll be back in 2012.

  • Andrew Heiss

    A cheat sheet and final predictions would be excellent

  • Behnam

    I’m interested in the role of geography and in why some states violate the geographic trends.

    The blue and red states are geographically contiguous except Colorado and New Mexico on the one hand and possibly Florida on the other hand.

    Demographically speaking, what makes these states different from the states surrounding them?

  • JJtw

    Just a simple request, really. A simple ranking of all of your states in decreasing order of probability of Obama winning, preferably in Editgrid or Google Docs format. Can that be done?

  • Sam Wang

    JJtw – If you want this sooner rather than later, simply look over in the “For fellow geeks” link. It’s all there. In particular, a file called stateprobs.csv has the percentage probabilities of an Obama win (first item on each line).

  • Bill

    Why isn’t the Indian vote listed as having some power? Is it because a win in IN for Obama would mean a landside? I’ve been working on getting out the vote in IN–my first such polical activity since being a college student in 68′. Obama signs are being stolen in my area in northeast IN, but now that are starting to show.

  • gprimos1

    Dr Wang,

    Let’s see if anyone can follow my garbled request. Assume that the polls are biased towards Obama and that the ones that show a much closer race represent the true proportion. I would like to see what the EV estimator would have shown over time if these polls were used to estimate the proportion of each state.

    It seems like there are two ways to do this. One can identify those polling agencies with a neutral or pro-obama bias relative to the other polls and simply exclude them from the analysis. Alternatively, one could take a lower percentile (33% or the min of 3) instead of the median (50%). The second has the advantage that is probably easier to do in your code.

    This would provide a “worst case” EV estimator. If the 95% CI did not contain 270 then Obama would be expected to definitely win. On the other hand, if McCain ends up winning (legitimately) then this EV chart would show the “true” history of events that lead to his victory.

  • DanM

    I second the request for a Sam Wang / Nate Silver celebrity death match, mediated by Steven Colbert, perhaps. Or maybe just a two-man bowling tournament.

  • David U

    It is possible to do some sort of contingency analysis, ala, in which you look at the probability of one candidate winning one state (e.g, Ohio) if he lose another).

    538’s analysis seems to assume independence of state voting, when in fact they are probably correlated, so good luck!

    So, if, for example, McCain loses Colorado, the loss is probably correlated with factors that are influencing other states, such as very high minority turnout. So the pure simulation model that Nate Silver uses probably overestimates the chances of a candidate winning one state if he loses another

  • Displaced Canuck

    As with many others, a cheat sheet is my first request. As well, I echo Sherry’s request for your analysis of Florida. This is for personnel reasons only, son is working for Obama in Florida and I want to know how elated or disappionted he is going to be after working 18 hour days nonstop for two months.

  • Bill N.

    I have a statistical question about the median EV chart. On there you have a 95% confidence interval marked. I have been wondering if that is a bootstrap type of confidence interval, and whether it takes into account the multiple sampling error values from the various polls from the 50 states. It seems to me that that any 95% CI you compute would need to take into account the fact that you have at least 50 sampling error values (assuming as single poll from each state), and almost certianly more than that since you are combining multiple polls from each state.

    Thanks for your answer. I have enjoyed lurking and reading this site for several weeks now since I found it.

    Bill N.

  • Mark S

    Unlikely – but for Tuesday early evening, how about a Virginia cheat sheet/projection based on the county-by-county vote?

    Some networks may be publishing county-by-county vote totals as they come in. If these numbers were crunched in comparison to the 2006 Webb-Allen county-wide vote, it would be relatively easy to project the final vote.

    Virginia is likely to be the first state to report – and (as written on if Obama wins Virginia, it’s over! I expect to be watching the Virginia vote totals come in, but with no idea whether they represent red or blue parts of the state. So I wish that someone would crunch the numbers and post the Virginia vote with a projection or adjustment based on which counties have reported.

  • LT

    Who should you follow, please, online (or possibly on cable) that will be first out Tuesday evening with accurate extrapolations of national/swing-state voting trends based on detailed results or exit polls from the states whose polls close first.

    In simple English: Where we will get the best read on the presidential and tight Senate races before 8 p.m. Eastern, and when?

  • William Shanley

    Hi Sam–Are precise dates available for the event-induced fluctuations shown on your EV estimator? Thanks for your genius work! William

  • ICDogg

    Let’s say you have a race (like NE Senate) where the data is all at least a month old.

    How much does this affect the confidence/margin of error of the projection?

Leave a Comment