We've moved!

Social Icons

twitterfacebooklinkedinrss feed

Sunday, October 24, 2010

Polls Drastically Diverge on SD Governor's Race

Someone is doing really bad polling on the South Dakota gubernatorial race. Republican-run Rasmussen Reports finds Republican candidate Dennis Daugaard remaining above 50% against Democrat Scott Heidepriem. The specific numbers: Daugaard 55%, Heidepriem 36%, undecided 7%. Those figures do show a slow, steady, but far from cork-popping uptick in Heidepriem's numbers. Team Daugaard has cited internal polling done by Glen Bolger of Public Opinion Strategies pegging the race at 58–30 in favor of their guy.

Democrat-commissioned and local RMA Research releases numbers finding Daugaard's lead has actually shrunk to six points. After the eyebrow-raising 45–32 Daugaard–Heidepriem margin found by RMA several days ago, Team Heidepriem now touts RMA numbers showing the race at 41–35, advantage Daugaard. Another poll from Sioux Falls-based newcomer Nielson Brothers polling finds the governor's race even tighter, 43–40 for Daugaard.

These polls appear to be looking at different realities. The differences from Rasmussen are enormous. In nine days, someone is going to be seriously egg-faced.
Some interesting details from Rasmussen:
  1. Republicans are more solid behind Daugaard (77%) than we Dems are behind Heidepriem (63%). Given that Republicans outnumber Democrats here, Dems can't win if they don't line up behind their guy.
  2. "Fifty-nine percent (59%) of voters who say the economy is improving support Heidepriem. Daugaard has the backing of 71% of those who feel it is getting worse." That seems odd: the lieutenant governor, the optimistic quasi-incumbent, gets more support from the pessimists. The Democrat voice of change wins a majority among those who think the economy is on the right track.
Update 14:35 CDT: The Nielson Brothers poll on the Congressional race finds Stephanie Herseth Sandlin leading Kristi Noem 41.5% to 40.0%. That's within the margin of error; it's also within the margin of Marking—i.e., Indy B.-Thom, who gets 1.8%. NBP finds 16.5% of the electorate undecided on the Congressional race.


  1. Republican run? Rasmussen is considered an independent poll by all serious prognosticators including Cook Political Report, Rothenberg Political Report, and Real Clear Politics. There are reputable D and R firms like Wilson Research (r) and PPD (d) which are considered reputable as well. But none of them "qualify" Rasmussen as a Republican firm. Nate Silver (Flemings stat go to guy) praised Rasmussen for his accuracy in 2000, 2002, 2004, and 2006. In 2008, he criticized Rasmussen for not catching the Obama enthusiasm gap until late in the election. Early this year, Silver criticized Rasmussen for "over-estimating" the GOP enthusiams gap but has now admitted Rasmussen was ahead of the curve and correct.

    But if you insist on the "adjective" (which hints to be a pejorative), why don't you ascribe the same "adjective" to RMA? Because they work solely for Democrats and they released their methodology (ala Wilson and PPD), they are much more closely represented as a firm like PPD than Rasmussen.

  2. I thought I was being equanimous in my description. Republicans run Rasmussen. Nonetheless, their numbers are better recognized by various press agencies. RMA appears to be run by Dems; as I said above its work on the above cited polls is certainly paid for by Dems. David Montgomery says that NBP appears to be run by brothers on opposite sides of the aisle. Do note that I'm not trying to portray the Dem polls as more reliable than Rasmussen. I simply wondering which outfit is getting the state of the electorate so wrong.

    I would think that "Democrat-commissioned" carries equal hinting-pejorative power to "Republican-run." But the floor is open for other interpretations.

  3. I probably over-reacted to your earlier disparagement of Rasmussen.

    With regard to the weight given to independent polls and those commissioned by campaigns, Nate Silver has an old article on this matter either at fivethirtyeight.com or nytimes.com under fivethirtyeight. Basically, he doesn't give any weight to "commissioned polls" regardless of the pollster.

    While I don't recall if he discusses why "commissioned polls" are not reliable for ascertaining the status of the race, I'll offer mine. Commissioned polls are principally valuable to a campaign for determining messaging.

    They don't commission a poll to feel good (if ahead) or bad (if behind) but find out what might either build/protect their lead or find strategies that might get them ahead.

  4. No sweat: I'm backing away from the skepticism I expressed earlier this year about Rasmussen. If Heidepriem wins or even comes within ten points, I reserve the right to revive that skepticism.

    That makes sense: a campaign doesn't need to pay money to feel good about itself. It pays money to gain competitive advantage, like a business. But what advantage do you get paying for a poll that gives crappy results? Are the commissioned polls themselves all just testing marketing messages on the phone?

  5. I'll offer thes as possible things to consider when looking at a Rasmussen poll. Check me if I'm wrong on any of these, Cory and Troy.

    1. There is a built in MOE of 5% (I'm rounding up from 4.5. of you prefer to round down, use 4%... or just go with the 4.5 if it makes a big diff to you.)

    2. I'm pretty sure Rasmussen doesn't sample "cell phone only", respondents, just people with land line phones. So he's probably missing some socio/economic/demographic age group there.

    3. I'm also guessing that Rasmussen doesn't sample the Indian reservations.

    4. Finally, due to the small sample size, I'm thinking Rasmussen may not have captured the difference between SD's "urban" vc "rural" vote, or the difference between East and River voters.

    5. I think the Rasmussen could be missing some accuracy due to failing to recognize a voter segment that may not conform to their idea of "likely voters." I think there may be a pretty good chunk of people who are going to vote on issues this year (secret ballot, marijuana, and smoking ban) who don't usually vote on candidates.) Let's call this the I&R voters.

    Those in combination could produce results markedly dissimilar to the Rasmussen data.

    If I were a pollster, I would be taking all of those things into account. And I think some of them are. And I'll go a step further and say shat I doubt if any of those polls have been made public.

    We'll have to wait anther week or so to find out.

    Meanwhile, you can probably get a few clues as to what they night say by looking at the content of the ads people who know what they are doing are running.

  6. Man, I typed that bad. Sorry. I'm not gonna clean it up, even though I accidentally typed "shat" up there. I'm laughing too hard at myself to go up and repost it. Please forgive. ROTFL.


  7. First, a general comment. Pollsters usually poll a number of people with the statistical goal of being with a 4.5% margin of error 95% of the time. This looks like a wide spread however this same MOE is within 1% about 75% of the time (bell curve). Pollsters like to hedge their bet by talking about the widest MOE.

    Second, they construct a demographic profile of a populace based on party registration, age, likeliness to vote and in some cases economic (but not usually in these small samples).

    Third, they ask a certain number of voters for their views and adjust the results for the voter profile. For instance, if they only ask 10 people from the age of 18-24 but the profile should have been 20, they weight those results.

    Fourth, there is 9% undecided. They can break 80/20 one way or another or virtually split. Underdogs always hope for a big movement their way. Favorites depend on at worst a split.

    Does this seem imprecise? Yes. Is it? Not really because the modeling usually cancels out the outliers. But to some degree, this is factored into the margin of error.

    1) The reality that some people aren't polled is correct. I.E. cell phone users. So far, conventional pollster wisdom is this profile is captured in the demographics. I disagree with Bill there is a major demographic missed.

    2) I think the biggest exposure to Rasmussen (and all polls) is the unique factors of a race outside of the two campaigns (i.e. the impact of ballot questions on turnout profile and turnout itself). To the extent they aren't captured in the likely voter model, this can contribute to poll results which differ from ballot results. But again, I think they probably just make the margin of error more possible.

    In the end, you can't judge a pollster based on just one result. You have to look at their total body of work. You will note that Zogby used to be considered by everyone as "reliable" but no longer. While he says he changed his business and does less political polling, the reality is he had some rather bad election cycles (independent pollsters like Rasmussen and Zogby keep the lights on not by their political polling but what they do for businesses, i.e advertising focus groups and product launches. Getting a reputation for bad political polling really impacts their core business).

    SurveyUSA and Quinniapac used to be considered reliable but they too have had relatively bad results. Rasmussen and PPD (identified Dem firm) have risen in reputation over the last few cycles.

    To be continued

  8. And my continuance:

    Rasmussen is polling over 100 races regularly. Statistically, he will miss 5 races and be pretty accurate on 95. My point is Rasmussen could miss a race by 10% either way and it doesn't mean his methodology was wrong. Or he might hit a race exactly on and it doesn't mean his methodology is right. But, when Nate Silver analyzes him over 100 races, we will have a good idea on his methodology.

    Nate Silver (who I've come to love. Thanks Bill) looks at all these factors and assigns a probabability of the polls correctly picking the winner without regard to margin (which is what your comment belies with regard to SH coming within 10 points).

    In the South Dakota races, he gives Noem a 78% chance of winning and Dauguard 99%. Note my first paragraph where the MOE of error is 1% in 75% of the races under this methodology. Silver's odds on the Noem/Herseth race mirrors that because of the closeness of the polls. Similarly, the odds of a 19% spread being wrong is probably 1% as represented by his prediction on the Governor's race.

    Where Bill and I really agree is there is way too much talk about polls in individual races. I know it is a cliche' but the only poll that counts is election day and the campaign continues right up to the the last day.

    Frankly, I'm not pleased campaigns release internal polls. If you have the nomination, you get to fight to the end and need no further justification. And discussion of poll results distracts from the issues.

    I don't think you or Pat should be substituting poll discussion with issue discussion.

    I don't think media outlets should do polling.

    What is important is what polls tell campaigns. Without seeing the polls, I can surmise pretty accurately what Noem/Herseth/Dauguard/Heidepriem are seeing in the poll based on their advertisements. And, Bill does too.

  9. Points all well worth reading, Troy. And while I will likely continue o get caught up in the hype, I agree that we should talk polls less and policies more. At best, polls tell us the percentages of people who have already been swayed by the great messy combination of policy, marketing, image, prejudice, and bull. The number of people who think someone is right for the job does not establish who is right for the job.

    Troy, I particularly like the point you make that campaigns only contribute to the distraction by releasing the internal polls. When I ran cross country in high school, I never looked back, because looking back wastes time. If you're running as hard as you can, there's no reason to look back. Same for campaigns: it doesn't matter whether you're in front or behind; just run like heck!


Comments are closed, as this portion of the Madville Times is in archive mode. You can join the discussion of current issues at MadvilleTimes.com.

Note: Only a member of this blog may post a comment.