Is Rasmussen Cooking Its Tracking Poll Topline?

by: Chris Bowers

Tue Oct 28, 2008 at 16:45


This will probably cause me to lose my subscriber account to Rasmussen, but there is a major discrepancy in the internals of the Rasmussen tracking poll today that I have to point out. Simply put, no matter how I work the number, Obama has a larger lead than the 51%-46% published on their website today.

Here is the problem:

  1. The partisan weighting targets for the Rasmussen tracking poll are currently 40.0% Democratic, 32.8% Republican, and 27.2% Other.

  2. According to the subscriber page on Rasmussen's website, the partisan internals for the Rasmussen tracking poll are currently as follows:
    • Democrats: Obama 88%--11% McCain
    • Republicans: McCain 87%-10% Obama
    • Other: Obama 50%-44% McCain

  3. If these partisan weights are combined with these partisan internals, the result is Obama 52.1%--44.9%, or 52%-45% instead of the current 51%-46%. McCain. Even if all the partisan weighting and partisan internal numbers were rounded as favorably to Obama as possible on the subsriber page, Obama still leads 51.54%--45.43%, which is still 52%-45%.

The math I did in these calculation can be seen here. No matter how the rounding is done, the campaign is 52%-45%.

Unless I am missing something, Rasmussen reported today's topline numbers to be closer than they really were. Their own partisan internals say the campaign is 52%-45%, not 51%-46%. As far as I can see, either they are cooking their numbers, or they are feeding their subscribers false information. Given that I have paid $80 for this service so far this year, either way, I am really irritated.

Chris Bowers :: Is Rasmussen Cooking Its Tracking Poll Topline?

Tags: , , (All Tags)
Print Friendly View Send As Email
Is Rasmussen Cooking Its Tracking Poll Topline? | 25 comments
No way it could be 51.4 to 45.5? (0.00 / 0)
If you play around with rounding up or down the D/R/I numbers you should get there eventually. Anyway, it doesn't really matter - it must be steady else he would be saying so.
Chris did the math (0.00 / 0)
if you follow the link to the pdf you'll see even if you assume Obama had the rounding advantage on all the internals it still isn't possible to get to Rasmussen's topline numbers.
[ Parent ]
No, it doesn't work (0.00 / 0)
Check the PDF. I did the following rounding the the partisan weights:

D: 39.950
R: 32.849

And then I did the following rounding to the internals:

Ds: O 87.50%-11.49%
Rs: M 88.49%-9.50% O
OIs: O 49.50%--44.49% M

Even then, it is 52%-45% Obama. Something is wrong.

[ Parent ]
Trial and error (0.00 / 0)
There are numerous different ways beyond that. Round to favor McCain. Round up or down within each subset. Play around with it some more. There must be an explanation.  But, again, is it really that important? Pretty obvious to me that much of the Ras movement from day to day is down to rounding up or down. Obama +6 with indies was all I needed to know. Though it could only be 5.5! :)
[ Parent ]
Give me another way (0.00 / 0)
I added a full tenth to McCain's margin in each crosstab. If you have other ways of working on it, let me know.

Either the topline, or the partisan internals, are wrong. They don't match up.

[ Parent ]
Well (0.00 / 0)
I can get to their top line numbers (whether or not through a plausible or rigorous method).

Use your math with the bottom table (the "most friendly McCain" rounding for the partisan breakdowns and number. Then eliminate significant figures following the partisan changes (so 35, 3, 13 for Obama instead of 34.9563, 3.1207, 13.464). Add up those numbers and you get 51-46.

Now that is a pretty lame way to do it, but I bet different methods of rounding can easily move any poll a point or two. Also note that for this to happen you'd have to have that very pro McCain "Other" rounding. That factor has to stay below 13.5 and your most McCain friendly scenario is at 13.464.

It's doable...  

[ Parent ]
Since you're a subscriber... (4.00 / 3)
And also a well-known blogger, surely you could contact Scott Rasmussen and ask for an explanation?
I gotta agree with leshrac55 (4.00 / 2)
Before calling Rasmussen out like this in a very public forum, you should have asked them first and given them a chance to explain themselves.  Maybe there's an explanation that you didn't think of.  Or maybe you caught an error that they didn't notice, and they could have issued a correction.

Of course, if you did try to contact them and they told you to f*** off, that's different.  But you didn't say one way or the other.

[ Parent ]
Good catch (0.00 / 0)
I guess, though, that all this so called tightening nonsense is occurring in heavily populated states because I haven't seen a single state poll today that indicates any such narrowing on a consistent basis.

If anything, the opposite is true if GA and AZ are any indication.

Still think these numbers... (0.00 / 0)
May be reflective of the state of the race last week, when Obama was surging.

On the other hand, perhaps it's the trackers that actually lag, and the state polls that are ahead of the curve.  I just sort of assumed the opposite, but who knows.

[ Parent ]
I don't think so.... (4.00 / 1)
Rasmussen finished second in the last election in terms of it's accuracy. Also, and more importantly, Gallup's traditional likely voters has the race at 2 points; GWU has it at 3 points; IBP has it at 3 points; ARG has it at 5 points.  The polls are all over the place, and I don't think it's because any pollster is cooking numbers.  It has to do with how they are modeling likely voters (which can shift from day to day). Pollster.com has excellent analysis of the polls.  
This has nothing to do with accuracy (4.00 / 2)
this has nothing to do with Rasmussen being an accurate pollster. This is about Rasmussen either giving its subscribers faulty partisan internals, or giving everyone a faulty topline.

It is about their own numbers not matching up with each other, not about their accuracy in predicting the election.

[ Parent ]
That would be correct (0.00 / 0)
Push 'em on this, Chris.
[ Parent ]
Could they have used the old partisan weights? (0.00 / 0)
I'm assuming the potential for a simple mistake -

Based on the old partisan weights
D 39.7%
R 33.0%

I get a - possible - O51.4 / M45.5
based on the worst case w/r/t rounding.

Yeah, that seems to be a bit of a stretch, but not impossible.

Perhaps (0.00 / 0)
That could be it. It is also possible they used the old partisan weights for one or two of the days, and the new ones for one or two of the days.

That is about the only idea I have had so far.

[ Parent ]
The number of errors they constantly make updating the website (0.00 / 0)
That wouldn't surprise me.
[ Parent ]
Here's how they did it (4.00 / 2)
They rounded each subset before they totaled them up.  

Using the most favorable rounding possiblities in each category would yield the following:

O votes 35 D, 3 R, 13 I = 51

M votes  5 D, 29 R, 12 I = 46

Not sure why they would round before aggregating, but that appears to be what they did.

But even then (0.00 / 0)
They are feeding their subscribers bad partisan internals. Either way, I'm annoyed.
[ Parent ]
Err... (0.00 / 0)
I don't like it either, but they aren't bad they're just rounded when you don't think they should be.

And Rasmussen clearly has issues with rounding.

And Mr. Rasmussen said that averaging polls to one-tenth of one percentage point, as the sites often do, conveyed a false precision.

[ Parent ]
Why is he doing it then? (0.00 / 0)
Four years ago he was reporting to one decimal point.
[ Parent ]
Party ID bigger sample (0.00 / 0)
The party ID is a bigger sample than the tracking poll, so it may make sense to post an extra decimal point there but not elsewhere.
[ Parent ]
My complaint (0.00 / 0)
I am not sure that I agree that they are feeding their subscribers bad partisan internals.  I think you have more access to data, which is what you signed up for, and there is nothing to say that that data is bad.  

The problem with what they did is how they appear to have used that data.  By rounding before aggregating, they introduced additional error into the model.  How does this affect the MOE in the poll?  Does it go from +/- 3 to +/- 3.5, for example?  Has that been calculated into their report?

I think it is intellectually dishonest to do it, and, again, I am not sure why they would take the extra step to do it as simple rounding in Excel would not cause it to happen.  You have to enter them in as new whole numbers to get to their result.  I think it is a fair question to ask "why" they are taking this extra step.  Is it, for example, being done to perpetuate a myth of a tightening election or to advance a particular party's agenda?  

Ironically, in this case, I think that close polls will actually help mobilize marginal Democratic voters that may be growing complacent, so I am not sure who this would help in the end.  

[ Parent ]
I suppose the premature rounding theory could be correct (0.00 / 0)
I recall a study of scientific journals that showed that certain last digits in reported numbers were favored (probably 0, but I'm not sure), which is a sign that rounding occurred inappropriately in preliminary steps.  

New Jersey politics at Blue Jersey.
Wow (4.00 / 1)
Does someone ever have too much time on his hands.
It's easy to get funky averaging results (0.00 / 0)
I didn't actually go through your numbers, but...I've played with averaging polling results before.

Sometimes you have to go back to the raw totals to get your best results. This is particularly true when you are rotating data from one set of cross-tabs to another.

I'm not saying you did anything wrong, just that working without the decimal point or without the numerical totals can lead to layers of inaccuracy.

Is Rasmussen Cooking Its Tracking Poll Topline? | 25 comments
USER MENU

Open Left Campaigns

SEARCH

   

Advanced Search

QUICK HITS
STATE BLOGS

Home | RSS/XML Feed | Register | Contact

  © OpenLeft 2007-2010, Powered by SoapBlox

Powered by: SoapBlox