Skip to main content

View Diary: Electoral Vote Math: Obama 193, Romney 0 (106 comments)

Comment Preferences

  •  I also enjoyed the post and agree with your (7+ / 0-)

    analysis as far as it represents the "probabilities" of certain combinations suggested in the current data-set. suggested by the current state of Nate Silver's database, and modeling.  

    And, not referring to this author, or the use of "probability" here, which is defined as contingent upon the existing dataset, and not a forecast, I think I share what I think are some of Nulwee concerns with what appears to be a misunderstanding by some here, and other places, of what Nate Silver's means by a "73%" chance of an Obama victory.  Buried in Nate's explanation are important caveats he mentions, but seems to be widely misunderstood, one of which is "given the data, and model assumptions as represented, in these polling datasets, today."  

    I don't believe Nate claims he has any way of forecasting, what the future economic reports will be in coming months, nor how polls could change in response to changes in polls due to the 2-1 advantage the GOP has in expected advertising budgets, debate performance, an economic collapse in Europe, a war in the Middle East or a unpredictable gaffe.  

    Anyone claiming to forecast what the "actual" probability of a victory for either side, would have a lot of mathematical explaining to do.

    So, I'm delighted that Nate gives an aggregate forecast, and find them very useful, especially in communicating in a quick, simple and easy to think about way, the directions of trends in data analysis, but they can make no valid mathematical claim of being able to predict the real probabilities, of the component assumptions, such as the stock market, or what the polls will be like a two months from now, things that are inherently unpredictable.

    So, the word probability, as Nate, and others use it here, are not the same kind of meaning the word has when a mathematician tells you the probability you have of winning the lottery, or even "developing cancer" if you smoke, (which is already crossing of into "issues.""

    I thought the "NowCast" nomenclature was an innovative way to explain this difference, but it appears Nate uses that terminology for a slightly different purpose, I haven't had an opportunity to study the model, and assumptions in detail.

    My understanding is that there is some emerging academic evidence that betting markets, such as Intrade, do a better jobs than experts, or expert models at forecasting some kinds of events that experts, or expert based modeling such as Nate's, despite the fact that they lack the analytic sophistication, subject to a lot of qualifications.  

    Intrade's markets, which have their own "unique way" of calculating "probability" suggest Obama has about a 56% probability at this stage.

    But, we have no way of forecasting what the probability that the devastating drought in the agricultural sector may lead to a devastating plunge in fall economic reports, or that a $100 million advertising blitz might moving public sentiment 5% in Romney's direction, in the next month.  If this happens, Nate's mid September models could end up "predicting" Romney has a 55% change of winning, and we have no valid mathematical way of figuring out what the probability that that will occur.  And, I strongly suspect Nate would agree to this, because he does not make this kind of claim.

    BTW, I'm no criticizing the author in any way, for saying "given the current state of the underlying state by state poll averages, in this database, and the given assumptions, that they will not be changed by turnout dynamics, here how many paths that exist that mean these combinatorial probabilities.  This is again a different use of what the word means.

    We need to be on guard that we do not become either complacent, make faulty judgement on how we allocate our resources  based on misunderstanding of what some of these concepts mean.  

    The means is the ends in the process of becoming. - Mahatma Gandhi

    by HoundDog on Sun Sep 02, 2012 at 11:10:54 AM PDT

    [ Parent ]

    •  This is a very well-reasoned post. (2+ / 0-)
      Recommended by:
      HoundDog, Shakludanto

      But I'd point out that there is a flaw in combining probabilities that are not independent. Consider this:

      Suppose a poll in Indiana unexpectedly showed that the state was closer to a tossup? Would you not think that whatever factors caused this shift would also affect the likely outcome of Michigan, Wisconsin, and Ohio? IOW, a slight drop in the probability that Indiana would Vote for Romney be cause for grave concern in the Romney camp, even though they might still consider Indiana a safe red state.

      I do recognize that you understand this, but many people don't understand how tricky it can be to combine probabilities.

      Occupy is the symptom. Fundamental reform is the cure.

      by Tim DeLaney on Sun Sep 02, 2012 at 03:56:30 PM PDT

      [ Parent ]

      •  Excellent point. Also, the multiple regression (2+ / 0-)
        Recommended by:
        Tim DeLaney, Jerry J

        mathematics apparently used for some of Silver's, and most other economic modeling, are only valid if, as you point out, the variable are independent, and the structure of the system doesn't change across the time period of the empirical data used, and the period of the forecast.  In this case, the only things we can say with nearly 100% certainty is that neither one of these assumptions is true.

        This is a problem with vast numbers of regression based studies, used in corporate marketing, strategy applications, etc, and we do it anyway, because it sometimes it helps discuss and understand data, especially the directions of trends, or changes, as done here, and what else can we do, but we need to remember we are violating the foundation assumptions of our own mathematics, and not overdraw conclusions.


        The means is the ends in the process of becoming. - Mahatma Gandhi

        by HoundDog on Sun Sep 02, 2012 at 07:54:42 PM PDT

        [ Parent ]

    •  thank you for your thoughtful comments (2+ / 0-)
      Recommended by:
      HoundDog, blueoasis

      I agree that Silver's use of the word "probability" is unfortunate, primarily because people will naturally assume that it has a meaning they are familiar with-- say in poker, or even weather predictions.

      The purpose of this diary was to take Silver's numbers-- that I assume are generally reasonable estimates of Obama's chances-- and see where they lead.

      But even if it is "193 to 0" today, we've still got a lot of work to do from now until Nov. 6. And not just in the Presidential race, but also, critically, in electing a bluer, better House and Senate.

      •  Exactly, I do the same thing and Nate Silver's (0+ / 0-)

        model, database, and columns are among my favorite.

        And, noted that I learned much from you post, and you acknowledge these points.  

        But, elsewhere on this site, you can find near celebrations, and trash talking about out 73% changes, by people who clearly do not understand the most basic issues with regard to the lack of underlying validity of the nature of the conclusions they are drawing.  Nate, correct notes many of these issues, but even he seems comfortable applying multiple regression techniques  across four decades of elections, when every basic mathematical textbook I've ever seen on regression note the techniques are only valid if the variable in the system are independent and the structure of the system does not change.

        Both, things we know for certain are happening in political systems and polling.  Nate Silver knows, this, and even notes that cell phones, demographic stranger caller avoidance, and other issues, have made "a mess" of polling statistics.

        I've gone ahead and presented simulations in professional work, with even vastly less valid data, and methodology than here, so I'm not being a purist.  But, I do this under the conceptual umbrella or "scenario analysis, models for learning, parameter insensitivity in well constructed dynamics simulation models with internal feedback loops."  But, explicitly warn against using these kinds of models for forecasting, or describe the output scenarios as predictive probabilities.

        The means is the ends in the process of becoming. - Mahatma Gandhi

        by HoundDog on Sun Sep 02, 2012 at 08:04:05 PM PDT

        [ Parent ]

    •  Yes.. And, it is still too early to forecast (0+ / 0-)

      given the closeness of the polls.

      You are completely correct that most people ignore Nate's warning about probabilities.

      Romney can easily win this by winning the following states:
      No Carolina

      Those are fairly good possibilities

      If he then adds Michigan and Virginia, that's it.  He doesn't need Ohio or Pennsylvania... or Colorado.. or Nevada..

      Add to that.. I don't believe Nate can predict voter turnout.  I predict high voter turnout for GOP candidates, and good turnout for Dems, but nowhere near the record numbers of 2008.  Youth vote will be much diminished.  I have no scientific basis for that though.

Subscribe or Donate to support Daily Kos.

Click here for the mobile view of the site