Princeton Election Consortium

A first draft of electoral history. Since 2004

Post-moocher bounce?

September 18th, 2012, 3:00pm by Sam Wang


I’ve written before about an unusual survey, the RAND American Life Panel. It allows the same people to be followed over time, thus allowing a look at individuals who change their minds.

Take a look at what happened the day that Mitt Romney’s remarks were leaked, in which he said that 47% of Americans were Obama supporters because they expected food, shelter, and medical coverage. Outside the confidence band. Zoot allures.

The net change day-by-day is the difference between the blue and red curves. All opinion shifts are potentially transient. But the fact that Romney-to-Obama has exceeded Obama-to-Romney for 10 days in a row raises the possibility that the DNC and its aftermath are moving opinion in a lasting manner.

Tags: 2012 Election · President

19 Comments so far ↓

  • Nicholas Warino

    Was not expecting a Zappa reference at this site. Well done.

  • wheelers cat

    I have decided I really like this poll. And its not because of my organic liberal tendency.
    Its because it exploits new media (email survey) and still retains randomness (shifting 1/7 of the respondents). I still haven’t seen anything on the selection of the “panelists”. I would like to know more about that.
    Mr. Freeland mentioned Nassim Nicholas Taleb, the black swan/tail-hedging guy in the comments to his post.
    I too am a believer in the great value of randomness.

  • Peter D

    Idea. 2 timeseries. Rolling 7day sum of OtoR less RtoO. Rolling 7day meta margin change. Scatterplot. ??? Profit.

  • Jacob Hartog

    Just to clarify–
    The graph shows the % who have changed their intended vote in either direction (or from undecided to decided) from the baseline they reported in early July?
    Are we seeing the same potential voters change their minds multiple times?

  • Olav Grinde

    @Wheeler’s Cat: “…my organic liberal tendency”

    Ha ha! That’s right up there with Stephen Colbert’s “…and we all know reality has a liberal bias.”

  • Peter D

    OK, I did the analysis myself… there’s a decent correlation, but it has not been a leading indicator so far this year. At best it gives a 1 day lead over the MM.

    scatter: http://i.imgur.com/3FwWz.png
    ts: http://i.imgur.com/goaEs.png

  • BigAngryBubba

    Edited for clarity. -SW

    I say, my dear fellows.

    You all seem to be a bit Red, or at least pink. Soon my fellow Americans will become wise to your shenanigans, at which time you will be well and truly licked.

    You in particular, Professor W., should ready yourself for a figurative tarring. The good souls on my side will see that you are well taken care of. A gulag might be overstating it a bit, but might I suggest the salutary effects of honest labor? Certainly intellectuals have little place in the America I foresee.

    Yours sincerely,
    Bubbus Iratus Maximus III

  • Brad Davis

    What about looking at the periodicity in the data in the form of auto-correlations. Since the same sample of 500 individuals are polled every 7 days there is an element of auto-correlation in the data. How about examining the changing the mean change over each 7 day block? e.g. 9/10 to 9/3, 9/9 to 9/2, 9/8 to 9/1, etc. Some days may be more resistant (or more prone to) shifting one way or the other. Or is this where some sort of Fourier analysis would be helpful? Where did you get the data Peter D? I could use DataThief to grab it from the figures, but I don’t have that program at work- it’d have to wait until much later tonight after I put the kid to bed.

    • wheelers cat

      I believe the 1/7th that respond are randomly selected.
      Its not just a rotation.
      Here is the methodology PDF.
      https://mmicdata.rand.org/alp/election/RAND-CPEP-methodology.pdf
      oh wow.
      “Representation is attained by using standard survey sampling methods such as random digit dialing and address sampling and not requiring prior internet access. RAND provides members who did not previously have internet with internet access.”
      Holy Markov chains!

  • Peter D

    @Brad Davis

    Meta Margin is column 14 here: http://election.princeton.edu/code/matlab/EV_estimates.csv

    Rand Shift data is here: https://mmicdata.rand.org/alp/showfile.php?p=shiftweek.csv&t=excel

    You may have some sophisticated filter, but just looking at the autocorrelations nothing jumps out.

    If I smooth the data in the way you describe, then it appears that Tuesdays are the least likely to be pro-Obama and Sunday most likely. But it looks like noise – not enough datapoints.

  • Alan

    I’m still a bit confused by this graph. It seems to center around one percent. Does this indicate that each week one percent of respondents have changed their mind since the previous week? Since the poll first started? Am I anywhere near correct?

    Help please!

  • Matt McIrvin

    Man, the commenters on 538 are really starting to freak out over individual polls ticking down. Must be that time of year.

    • wheelers cat

      I dunno. The forecast and the nowcast seem to be diverging in Nates model. Is that reverse Central Limit Theorem?

  • Ralph Reinhold

    It is interesting to look at all of the tabs on the RAND report. The ‘intent to vote’ is interesting when compared to this. In some cases, changes in voter preference have a counter effect on intent to vote resulting in little change in likelihood of being elected. The moocher bounce is not one of them.

  • Barb

    I remember last presidential election at 538 as time was running out and the media was frothing, someone finally cried, “Hold us Nate, we’re so scared!”

    Cracked me up, partly because I felt the same way. Now I have nerdy, soothing data from Sam when I inadvertently catch too much other media and need to chill.

  • Susan

    Note that this is not a day-to-day plotting, but rather a rolling cumulative 7-day time series.