EXIT POLLS: Want to debunk rBr?

[Bill Bored]

OK geeks -- you want to debunk rBr?

Page 54-55 of Mitofsky's Jan 19, 2005 report
<http://www.exit-poll.net/election-night/EvaluationJan192005.pdf>
lists refusal rates by state, and last time I checked, states
came in 2 colors, red and blue, right?

Even if some of these are in dispute (purple), or are swing
states, such as NM, NV, IA, OH, WI and PA, and who knows which
others, there are still some states who actually are able to
distinguish between the 2 parties and vote for the one they
like by overwhelming margins.

So if rBr is a fact, you'd expect there to be higher refusal
rates in the Red states and lower ones in the Blues, right? If
not, there would be no such correlation, or a negative
correlation.

So has this been looked at yet or what?
(Maybe I didn't get the memo.)

Here are the refusal rates by state:

All Precincts	35.8
	
Alabama	        31.9
Alaska	        36.9
Arizona	        28.4
Arkansas	31.1
California	35.9
Colorado	34.9
Connecticut	37.5
Delaware	25.8
D.C.	        30.8
Florida	        40.3
Georgia	        27.8
Hawaii	        34.5
Idaho	        30.8
Illinois	37.8
Indiana	        41.3
Iowa	        38.2
Kansas	        27.6
Kentucky	38.6
Louisiana	40.1
Maine	        30.8
Maryland	33
Massachusetts	30.6
Michigan	39.4
Minnesota	40
Mississippi	38.9
Missouri	44.1
Montana	        29.7
Nebraska	25.1
Nevada	        34.9
New Hampshire	46
New Jersey	29.8
New Mexico	32.3
New York	33.5
N. Carolina	36.5
N. Dakota	27
Ohio	        42.5
Oklahoma	34.5
Pennsylvania	41.3
R. Island	42.2
S. Carolina	28.8
S. Dakota	35.6
Tennessee	24.6
Texas	        28.6
Utah	        32
Vermont	        38.8
Virginia	35.3
Washington	38.8
W. Virginia	43
Wisconsin	35.6
Wyoming	        26.1

Of course we should weight these by population and also by
vote margin. In other words, the states with the highest Kerry
and Bush margins (partisanship) and the largest populations,
should have the highest weightings. Since we suspect all kinds
of election fraud, and we want to correlate with Red and Blue
states, I'd leave out any state with a victory margin for Bush
or Kerry of less than 10% or so. What do you think?

Anyone want to take a crack at this, or is it already out
there somewhere?

[autorank]

1. I think you missed the memo.

"The EAST was the rBr BEAST"

I believe this will save you or someone a lot of work.

[Bill Bored]

2. And that thread shows actual refusal rates as a function of state

partisanship WHERE exactly???

[autorank]

3. Don't know if the data's there. I just remembered the thread.

[TruthIsAll]

4. rBr Debunked: see USCV Simulation AND the Exit Poll Response Optimizer.

Check your data. The avg response rate was 53%, according to E-M.
Don't leave out ANY state.

http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x379134

To match E-M data, total Alpha (K/B) = 1.15 minimum:
1.17 in non-Kerry precincts and
1.07 in Kerry precincts.

In other words, Bush voters did not respond in Bush precincts, but they did in Kerry's?

[Bill Bored]

5. I'm talking about the 35.8% REFUSAL rate.

If you have reluctant responders, it means they REFUSE to take the poll. I don't care about completion rates per se.

On edit:
Having said that, there is also the Miss rate to consider.
100% attempts = Completion rate + Refusal rate + Miss rate

If the Miss rates confound either the Refusal or the Completion rates, then you have to account for them or ignore them.

So use Completion rates if you want, but intuitively, the Refusal rates seem to be the key since that's what rBr is supposed to represent -- higher refusal rates by Bush voters.

You can find Bush voters in Red states, and we have actual Refusal, Completion and Miss rates for every state. You can weight the states by population and if you think there's fraud, you really do have to throw out the 50/50 states because you don't know if they're vote count is correct or not. But this point is debatable I guess.

Note: I'm talking about using actual data by state -- not any kind of simulation or average -- and correlating it with the actual partisanship of each state to see if there were more non-responders in the Bush states or not (weighted by population).

I don't think this has been done yet, has it? Not simulations but the real thing?

Note also that based on the real data, alpha was NOT correlated with precinct partisanship.

What I'm suggesting may be a way of cross-checking that.

[andym]

6. Any explanation for the wide variation in refusal rate in New England?

A quick perusal of this data, shows a very wide variation of refusal rates in New England, a region I am familiar with:


CT: 37.5
MA: 30.6
ME: 30.8
NH: 46
RI: 42.2
VT: 38.8

Between NH and MA, or NH and ME there is about a 15% difference.
While MA is very liberal, I think ME and NH are fairly similar in partisanship-- not completely sure though.

Also RI and MA are both strongly liberal, why is there a 11.5% difference in refusal rates?

This is quite weird data. On the surface, it looks like this variation cannot be explained by rBr.

But, what does it mean, if anything?
Any explanations? Why are there such large variations at all?

[Bill Bored]

7. Well, that's what I'm proposing we look at

by doing a scatter plot. Refusals on one axis and Kerry or Bush vote percentage on the other, by state, weighted by population.

This way there would be 51 data points representing various levels of partisanship, based on actual data, correlated (or not) to refusal rates.

I agree that it looks like there are some inconsistencies. The question is what do they prove or disprove, and also whether or not this has been done before.

The whole thing with alpha and logn(alpha) frankly gives me a headache! Alpha is just a ratio of response % by candidate. It had no correlation to precinct partisanship. This in itself doesn't prove that alpha had any particular value (response bias), does it? Just that it wasn't correlated to vote percentages. So where's the proof for or against rBr in that? Someone correct me if I'm wrong.

Now on the other hand, if we knew that there were greater refusal rates in either Bush or Kerry states, that might tell us something about which responders were or were not reluctant.

Again, if this has been done already, using ANY actual data, please show it to me. Every time I read the various papers on this subject, I find something I missed before, and someone might have done this analysis already and even published it. I just don't recall seeing it.

[Melissa G]

8. Bill, I think this is something that i have being doing research

around. There is the problem of where to get the partisanship data from... As you know there is no national figures for partisanship and no consistency for how it is reported state to state so it seems you would not have apples to apples data for comparison to the refusal rate? Am I correct?
I'm shocked I'm almost talking exit poll techie talk...all this research must be rubbing off..;)

[Bill Bored]

9. almost talking exit poll techie talk!

I hate it when that happens! If you start to grow horns, change the subject, see a priest or find some garlic right away!

Well, partisanship for the purpose of these sort of discussions is just the margin by which someone won the precinct or state. So we do have a lot of data on that, except of course if the results were totally wrong. But that's OK too because we're comparing exit poll results to election results with the assumption that ONE of them MIGHT actually be right. Big assumption there though, IMO!

[LightningFlash]

10. I fully support this juxtapose analysis.

I think it would tell us more than all of these other optimizers are able to tell us now. It would do more good to combine the number of refusers, rates of partisanship, and precincts into one optimizer done by TIA or someone else. This would show us just at what level county fraud was done and how it was accomplished.