The Bishop and the Butterfly: Murder, Politics, and the End of the Jazz Age

    There no mo no mo

    By Flavius on Thu, 10/21/2010 - 3:50pm |

    Nate Silver denies that there's any such thing as "momentum" that can be established by comparing successive polls.

    If Charlie Candidate is up 5% in June and again in July that doesn't mean he's probably going to be up in Aug

    "If anything there is a slightly negative serial correlation-that is a candidate who gains ground in the polls from Period C to Period B  tends to lose ground from Period B to Period A."

     

    Comments

    I read N. Silvers this morning and saw his comment.  It made me mad until my grandson explained to me that Nate is a numbers nerd.  He don't see past or anything else but the probablitity of numbers.  I am interested in Artie's take on this too.     


    by trkingmomoe on Thu, 10/21/2010 - 4:48pm

    Well . . .

    I'd venture to say that if Nate actually sees that particular statistical correlation in his computational data then it's there.

    I can only take Nate's word for it and I consider his word solid from past history. I would have to see the actual complete set of data to confirm what he's seeing. Although, in the article he does go into a indepth explanation.

    Now I suggest that folks read Nate's entire article in it's full context. Here is key for me:

    "They create the impression that — if the candidate has gone from being 10 points down to 5 points down, then by next week, he’ll have closed his deficit further: perhaps he’ll even be ahead! There’s just one problem with this. It has no particular tendency toward being true."

    From that, I get the impression that short term current trending situations in relationship to Nate's vast amount of past computational data does not support the momentum aspect.

    ~OGD~


    by oldenGoldenDecoy on Thu, 10/21/2010 - 6:05pm

    Nate does not suggest that "momentum" never exists.

    If one were to actually read the entire article, Nate qualifies his position with the following statement:

    "Although this analysis might seem modestly technical, the conclusion is really pretty simple. In general elections, the direction in which polls have moved is not predictive of the direction in which they will move.

    Thus, it is usually wrong to say that a candidate is gaining ground in the polls — present tense — or that her position is improving. Instead, you should say that the candidate has gained ground or that her position has improved.

    Now, I don’t mean to suggest that “momentum” never exists. Surely there are cases where it does. But it does not exist as a general rule on the basis of the polling — at least not in general elections."

    One has bother to go to Page 2 to find that.

    ~OGD~

     

     


    by oldenGoldenDecoy on Thu, 10/21/2010 - 6:16pm

    Poll reporting over at TPM always drove me batty, in that every new result was overhyped. If candidate X lost two percentage points in August from what he/she polled in July, that was a trend.

    You cannot deduce momentum from two data points, even if individual polls were reliable reflections of reality, which they are not. Every poll has a margin of error and, as I've pointed out here before, every 20th poll result can be expected to be an outlier -- i.e., exceed the margin of error. So if X leads Y in July by less than the MOE, and Y leads X in August by less than the MOE, and if in fact the total change for either candidate is less than the MOE, all you should ever conclude is that the race is pretty close.

    Nate Silver goes further, showing there's no statistical evidence of "momentum" from one time period carrying over into the next, even when you consider shifts in poll averages (which should be more accurate than individual polls).

    Momentum can be real, Nate concedes, as voters become more aware and candidates score more points or make more gaffes. Problem is, polls are too blunt a tool to detect it as it happens.


    by acanuck on Thu, 10/21/2010 - 10:04pm

     TPM polls drove you nuts too ... Eh?

     

    Here's a post I dumped over at Once Upon a Paradigm that pretty much sums up my outlook on the TPM polls.

     

    ~OGD~

     

     


    by oldenGoldenDecoy on Fri, 10/22/2010 - 3:27am

    I have a like/dislike relationship with Nate Silver. I think he is impressively technically proficient at the mechanics of extracting meaningful statistics from large sets of raw numbers and isolating signals within those datasets (meaningful ones as well as noise). He also has a functionally plausible methodology for the dicy practice of aggregating polls which use widely divergent question formulations and weighting methodologies that is solidly documented and seems to perform rather well. In that regard, I consider him second to none.

    I have rather less respect for his punditry and think to a certain extent he commands a respect in the latter based on the reality of the former which is unwarranted. He tends to present in the definitive - ignoring the fact that highly improbable is indeed possible and hiding his caveats well beneath the fold. IMO, this leads to an irresponsible cascade of people taking his top-line conclusion and running with it in subsequent coverage of his opinions while stripping critical caveats in process. While indeed this final complaint is certainly more the fault of those providing secondary coverage, it happens far too often to say that he shouldn't also take a bit more responsibility.

    But by far, the most useless analytic metric I have encountered is TPM's "Poll Average". (bet you thought I'd totally ignored your actual comment ... ha!) WTF? On that I have a very similar criticism to the one Silver is presenting here. Each poll is a snapshot in time. All averaging polls does is erase the one real value of polling which is to show where the opinion of people queried sits at any given point. Erase the actual number and then temporally disconnect it from any concrete reference point  ... and it becomes all but meaningless. They REALLY need to spend a bit more time documenting their methodology and explaining it's validity like Nate does. Because to this casual observer, it just seems like bunk.

    (BTW: Is "Once Upon a Paradigm" the site formerly known as "Once Upon a TPM"?)

     


    by kgb999 on Fri, 10/22/2010 - 5:05am

    Yes, they cut the umbilical and became Paradigm.


    by Donal on Fri, 10/22/2010 - 8:10am

    What drives me crazy about the poll analysis over at TPM is that they include Rasmussen.  Rasmussen notoriously skews right, and seems to do more polling than any other firm out there (at least if you get almost all of your polling data from TPM, as I do).

    Not being a numbers cruncher myself, it seems like the end effect of including Rasmussen is to have a disproportionate number of polls in your sample with a right-leaning bias.  Which, of course, leads to more "Democrats are Doomed!" headlines early in the campaign season, further dampening Democratic enthusiasm and aiding the Republican cause.

    I'm guessing the folks at TPM know what they're doing, but their inclusion of Rasmussen still raises questions.  Do any poll watchers here have any comments?


    by brewmn on Fri, 10/22/2010 - 9:22am