Ron Baiman: Sunmmary of the USCV/E-M and Liddle Debate

[eomer]

<note from eomer: this post by Ron Baiman (http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x371726#372424) deserves its own thread>

Dear Democratic Underground Readers,

I have asked to post on Democratic Underground. I hope this "summary of the debate" is useful.

The following summary was written as a response to a post complaining that USCV (www.uscountvotes.org ) was making a fuss about minor discrepancies, implying that the E-M case had been proven by the scatter plot of the WPE-index (Ln(alpha) – the “Liddle” index) shown by Mitofsky at AAPOR.

The response is, almost in full, as follows:

a) USCV initiated the study of partisan exit polling response rates in order to show that implausible partisan response rates are necessary to generate the aggregate tabulations released in the E-M report (see USCV March 31 updated April 12 report). The report develops a methodology to calculate Kerry and Bush voter exit poll response rates (K and B) which when multiplied by their respective reported vote shares (k and b) and added up (assuming negligible independent vote) give an overall exit poll response rate (R = Kk + Bb). Table’s 2 – 4, p. 11 and p. 25 of this report, show that implausible variations in K and B from “representative” (mean valued) precincts would be necessary to generate the mean WPE’s (E) and overall response rates (R) shown in the E-M report.
b) Elizabeth Liddle, who was an active participant in the USCV discussion list at the time discovered that when simulating the logged ratio of possible K/B values – based on the derivations developed in the April 12 USCV report - (Log (K/B) - the log was added – I believe at the suggestion of another participant on the list - to make the ratio symmetric), over partisan precincts, a noticeable “inverted U” asymmetric shaped WPE pattern emerged. This was particularly noticeable when simulating an (implausibly) high 2:1 K/B ratio (see Liddle figure 1 - displayed by Warren Mitofsky as part of his AAPOR presentation).
c) Liddle may have unknowingly reinvented the “wheel” algebraically but her index (the WPE_Index displayed by Mitofsky at his AAPOR presentation) is equal to Log(K/B) where all these terms are based on the derivations developed in the April 12 USCV report (see Appendix C –D in latest, May 12, updated May 21, USCV report). The importance of this is not only one of correct attribution of credit (appreciation of the value of an analysis should not be dependent on the conclusions that are derived from it!), but also to point out that USCV and Liddle have been looking at the same variables (partisan exit poll response rates) derived in a mathematically equivalent fashion.
d) E-M seems to be mistakenly, or deliberately, trying to create the impression that the Liddle analysis is based on some new, hitherto undiscovered “artifact” or “confounding”, that resolves the debate and shows that the mean rbr hypothesis can, after all, in spite of appearances to the contrary, explain the data. This, unfortunately, may partly be a result of the appearance of Liddle’s paper, in which she derives “alpha” (in a rather complicated and convoluted manner) apparently without reference to the equations for K and B presented in earlier USCV reports. (The obvious way to get alpha, once K and B have been derived, is in one or two easy steps as is shown in Appendix C of May 21 USCV report.) In addition, though she refers to the USCV paper in at least one footnote, Liddle unfortunately does not point out that the USCV analysis was based on these very same K and B partisan response rates patterns that she investigates in her paper.
e) In any case, based on her finding of the “asymmetric inverted U” pattern, Liddle came to an opposite conclusion to that of USCV. She surmised that this pattern indicated that the E-M data could have been generated by an unvarying mean partisan response “bias” which she defined as K/B.
f) How could she have come to such a different conclusion? The USCV reports used E-M reported mean WPE and R values to calculate K and B levels in different categories of precincts.. These K and B values diverged markedly in implausible ways across partisan precincts. Liddle, on the other hand, simulated (K/B) to get the suggestive pattern displayed by Mitofsky above. By looking at an exaggerated (by assuming a 2:1 K/B ratio equal to an alpha of 2) simulation of the K/B ratio she produced an “inverted u” shaped WPE graph that seemed to produce an “inverted u” WPE graph that seemed curved enough, and asymmetric enough, so that it appeared that it could approximate the know E-M reported WPE outcomes.
g) In response to Liddle’s pointing out this asymmetric inverted “u” WPE pattern, USCV added Appendix B (which appears in both reports) which derives this pattern from “Differential Partisan Response” (w = B – K). This appendix shows why the “inverted u” pattern appears, why WPE will be at a maximum perfectly competitive districts (k=b=.5), and why differential partisan response (w) will be equal to WPE (E), if k=b=.5 and R = .5. Appendix B also points out that mean calculations that were already done in the USCV (April 12) report show that the “inverted u” finding makes the jump in WPE to -10% in high Bush precincts even more implausible (as highly partisan precincts should have smaller WPE), and that to get the WPE’s for the other categories of precincts, the calculations in the report show that w would have to swing from 40% to an absolute minimum of 20.5%.
h) In addition, Appendix B shows that the “asymmetry” of the “inverted u” WPE curve – which gives a larger WPE in high Republican precincts (see Liddle Table 1 and Mitofsky presentation) that seems to be consistent the (much higher) WPE of high Bush precincts in the E-M data, is a mathematical result of linking an absolute difference (WPE) measure to a ratio measure (alpha). This “mathematical nit” cannot possibly explain the dramatic asymmetry in the E-M data (see the WPE’s generated by a constant alpha = 1.15 in Table 2, p. 19, May 21, USCV report). Moreover, if an absolute difference “differential partisan response” measure is used (w= B-K), even this small asymmetry disappears altogether. Only with highly magnified levels of Alpha (such as a 2: 1 ratio representing alpha=2) will this small effect look significant.
i) This debate with Liddle went on for some weeks. USCV members did further calculations based on means and medians with “alpha” (=K/B) and showed that using either means or medians “alpha” would have to range from below or almost equal to 1 in high Kerry precincts to 55-58% above in high Bush precincts to get the E-M reported WPE outcomes for representative precincts (see Table 1 in Appendix F, p. 19 of most recent May 21 USCV report.)
j) USCV showed that the overall response rate (R) levels that would be generated by representative precincts under the E-M “alpha” hypothesis of K =0.56 and B=.5 were mathematically infeasible in high Bush precincts and highly implausible in high Kerry precincts (R would have to drop to 16.9% - 29.2% in these precincts). Note that Liddle’s index is simply a ratio index that says nothing about overall response rates. Though K and B can be generated from r (=(B+K)/2) and w (see Appendix B) , they cannot be calculated from alpha (=K/B) alone. Another parameter like r is needed to get K and B and overall response rates (see derived equation in Table 4 of Appendix 4.
k) Liddle attempts to address this in her paper by noting that the E-M report states that the overall response rate (R) changes are not significant, implying that the bias (alpha = K/B) is all that needs to be looked at. But this skirts the issue. If a unvarying mean bias (mean constant alpha) hypothesis can only be sustained with radically divergent and implausible overall response rates, the hypothesis cannot explain the data, which as noted, show small (and possibly statistically insignificant) variations in response rates. USCV’s first report also showed that these small response rate variations contradict the E-M hypothesis.
l) None the less, Liddle was not swayed that a constant mean response bias (alpha) could not possibly (with any reasonable degree of chance) explain the E-M data. A key issue at this point was whether “aggregate” calculations using means and medians (such as the ones the USCV had used in its first report and in Appendix B) could adequately represent outcomes from randomly varying precinct level simulations. Recall that Liddle’s analysis at this point was based entirely on hypothetical simulations. In response USCV did some output simulations – trying to match E-M WPE and response outcomes with constant alpha. These also showed that matching E-M reported mean and median WPE levels, and over-all response rates with constant alpha was highly improbable to impossible (see Appendices G and H in the May 21 USCV paper).
m) Liddle remained unconvinced. She went ahead and published her paper, which she had already written about on several web sites. Her paper was hailed as holding the key to saving the constant mean bias hypothesis. The fact that it was based on the very same data (and pretty much the same analysis that USCV had be doing to show the opposite) was overlooked.
n) USCV felt pressured to respond to the Liddle paper by releasing the May 21 paper, and sending representatives to the AAPOR conference. This new USCV paper in addition to the calculations in k) above, also included an input simulation that showed that the E-M hypothesis could not explain the E-M outcomes, even under the most extremely (favorable to the hypothesis) precinct distributions (see Appendix I ). Moreover, these simulators (one of which is in completely transparent Excel form) have been put on the USCV website so that anyone can verify these results.
o) The Liddle and E-M hypothesis had been rejected at this point in at least three different ways.
a) Analysis of “representative” mean and median precincts showed that it was mathematically infeasible or highly implausible (Appendix F, May 21 USCV report).
i. Alpha’s necessary to produce the E-M data change by more than 62% (from mean calculation), or 52% (from median calculation – see Table 1, p. 18, May 21 USCV report).
ii. An alpha of 1.15 (representing an even greater “bias” than the 1.12 hypothesized by E-M and Liddle) is unable to generate the large E-M reported WPE values for high Bush and competitive precincts, and the small WPE for high Kerry precincts (from mean calculations – see Table 2, p. 18).
iii. Under the E-M hypothesis of K=.56 and B=.5 (so that alpha=.56/.5=1.12), the overall response rates (R) for high Bush precincts would have to be a mathematically infeasible -564.4% (from mean calculations) or also infeasible 423.3% (from median calculations). Overall response rates in high Kerry precincts would have to be a highly implausible 16.9% (from means) or 29.2% (from medians), see Table 4, p. 19.
b) “Output simulation” shows that with a high degree of certainty, matching E-M mean and median response rates, and overall response rates, requires significant unexplained changes in K, B, and alpha (Appendices G-H, May 21 USCV report, especially table on p. 20).
i. non-uniform mean alphas that change by at least 31% across partisan precincts.
ii. non-uniform partisan exit poll participation rates (K changes by at least 16%).
iii. High Bush precinct Kerry voter exit poll participation rates that are much higher than Bush voter participation rates (by at least 16% and sometimes up to 40% to 60% higher – p. 21).
c) “Input simulation” shows that it is impossible, under the most extreme favorable to the Hypothesis circumstances, to get E-M reported results from an E-M constant mean K=.56 and mean B=.5 hypothesis (Appendix I, May 21 USCV report). In particular this simulation shows that the E-M reported:
i. High Bush precinct mean WPE
ii. High Bush precinct median WPE
iii. Low Kerry precinct mean WPE
iv. High overall response rates in high (b>.8) and moderately high Bush precincts.
Are all unobtainable under the E-M hypothesis.

p) None the less, E-M, based on the Mitofsky AAPOR presentation and your comments, has embraced the Liddle analysis as providing conclusive evidence for the constant mean bias hypothesis. The key part of this argument appears to be the WPE by LN(alpha) linear correlation analysis presented by Mitofsky at AAPOR that we have been discussing. The question of how such a simple correlation analysis could trump the data already presented by USCV regarding representative precinct mean and median partisan and overall response rates across partisan precinct categories, and the more recent conclusive simulation outcomes, seems to have been lost or deliberately ignored.
q) It should be noted that the E-M report itself simply supplied tabulations and discussions of “factors” that could influence exit poll response and bias and an assertion of the E-M, that partisan response rate of K=.56 and B=.5 could explain all of the WPE error (p. 31 of E-M report). No solid statistical evidence (for example a multiple regression analysis actually showing that the factors can explain the WPE patterns, and that these factors result in .56 and .5 response rates, is offered to support this hypothesis. By the way, I agree with you that the term “reluctant Bush responder” is inaccurate. I was simply using the term as it has been coined by the media. One of the ways to get beyond the simplistic psychological metaphors, to the real factors that influence response rates, is to do the serious multi-factor analysis.
r) Mitofsky claimed, when I queried him about this at the AAPOR conference, to have done the regressions but not released them. This is doubly unacceptable! If they were done, the public has a right to see them! However, I am somewhat skeptical that they have been done, or at least done in a thorough and complete manner, as cursory analysis shows that K=.56 and B=.5 (generating an alpha of 1.12) could not possibly explain the relative magnitudes of WPE shown in the E-M report. See Table 2, p. 19, May 21 USCV report: alpha had to be increased to at least 1.15 to get WPE’s in range of the E-M data. It seems to me that this hypothesis, stated with such certainty on p. 31 of the E-M report, could not have come from an in-depth and serious statistical analysis by some of the “best analysts” in the country!
s) As I stated in my earlier post, the kind of tabulation, and now linear correlation analysis, that E-M has released to the public, would never pass muster as supporting evidence in any kind of serious academic journal (including one that I am an editor of). The gullibility of the media in support of the constant mean bias hypothesis without any serious evidence for it has been a travesty. The notion that “these things take time” etc. is also unacceptable. The credibility of our election system is an extremely important national issue – it should not take six months or more to provide a serious analysis (especially if some of the “nations best” analysts have been looking at it) of such an important issue. Moreover, there is no reason that private business contracts or personal confidentiality should trump critical public interest in this data. There are ways to release this data that protect confidentiality (as has already been done for Ohio). There is no sufficiently important or legitimate reason for E-M not to release the data and very good reasons, relating to a minimal sense of public responsibility and survey ethics, for E-M to immediately release the data without further delay. This is what I meant by “E-M needs to release the data”.
t) It may seem a bit beside the point, after all this to be debating whether a single linear correlation analysis of the E-M data is consistent with the non-varying mean exit poll response bias hypothesis. After all, the statistics (mean, median, and absolute value WPEs, and mean overall response rates) for the different precinct categories have been calculated and reported on. These are clearly highly divergent. Moreover, the influence of precinct partisanship has been eliminated from these data by calculating the direct K and B response rates (all done in the April 12, USCV report) and these show that implausible changes in K-B and K/B (the log of which is the Liddle/Mitofsky WPE-index) are necessary to generate these data from “representative” precincts. Why then should we be debating whether an insignificant linear correlation between this WPE-index and precinct partisanship shows that it unvarying?! We have already done the calculations and the analysis showing that this is not the case! For those concerned about “aggregation bias” in using “representative” precincts, we have shown that this hypothesis is highly implausible, if not mathematically infeasible, with precinct level simulations as well – see point o) above.
u) I submit that whether or not the scatter plot, considered as whole, produces a significant linear correlation or not, is under these circumstances, irrelevant. After all, a zero correlation can be produced with any number of non-linear variations. In this case the range of alpha’s (taking natural logs of mean and median alpha columns in Table 1, p. 19, May 21, USCV report) is (going from low Bush to high Bush quintiles), from means: - 0.0166, 0.1448, 0.1704, 0.1414, 0.4626, and from medians: 0.019, 0.137, 0.168, 0.141, 0.438. This would seem to imply a lot of variation and a positive correlation (with b). However, evidently, because of the very small sample sizes for the highly partisan precincts, the “inverted u” (not flat linear!) shape of the alpha from 90% of the data that is clustered in the less partisan precincts is sufficient to generate a flat zero correlation.
v) This “inverted u” alpha (not evident when drawing a straight line through the scatter plot) suggests that constant alpha is insufficient to generate the large WPE for competitive precincts (-8.5%) relative to the lower WPE levels (-5.9% and -6.1%) for the less partisan districts. This in it self may rule out unvarying alpha. This will depend on the significance levels of these differences – but given the very large sample sizes for these precinct categories, small differences in mean alpha levels (of +0.03 or so) are likely to be significant.
w) You make claim that the only unusual thing about the scatter plot data are four high Bush outliers in the high Bush quintile that are not offset by any high Kerry outliers, and what’s the big deal about four points, though you support investigating these precincts. First, it is curious to have only 40 high Bush precincts compared to 90 high Kerry precincts (measured by reported election outcome) in an election that Bush won by 2.7%? One would expect rough equality or a slightly larger number of high Bush precincts in a representative sample. This looks strange and may indicate that some other high Bush “outliers” have already been dropped from the sample. Second, even if for some reason there were less than half the number of high Bush precincts in the sample, four outliers represents 10% of a sample of 40. If 10% of all of the high Bush precincts in the country were corrupted, this could represent a very serious problem.
x) Moreover, all of the other ways in which these data are not consistent with constant mean alpha cannot be addressed by simply removing the four outliers. Whether or not the high Bush outliers are removed, the USCV reports have shown, for example, that the E-M hypothesis is not consistent with the high Bush median (which presumably would not be greatly affected by removing outliers) with the high Kerry mean, and with the high, and relatively high, Kerry overall response rates.
y) However, I reiterate, the basic point about the correlation analysis of the scatter plot is that it is wholly inadequate. Such an analysis “of the whole” will not provide detailed (or accurate if the variation in alpha is non-linear) information about what’s going on in the most interesting extreme partisan precincts where the constant mean bias hypothesis is really put to the test. As has been shown above, detailed analysis of the E-M constant mean response bias hypothesis breaks down in multiple ways particularly in these kinds of precincts.
z) The constant mean bias conjecture remains an unsupported (and largely inconsistent with the data that has been made public) hypothesis. Six months after the election, we still have no serious explanation for the large exit poll discrepancy. The shoddy (if I could borrow a term) and inadequate analysis (claiming for example that tabulations and linear correlation analysis are sufficient to support the E-M hypothesis) that has been released to the public has just deepened the uncertainty about what happened in the 2004 elections. I don’t see how this could be viewed as anything other than a national disgrace. The volunteer work of USCV, and other citizen activists who are deeply concerned about the credibility and/or integrity of our electoral system, and have refused to be satisfied with this pabulum, may, in fact, be the one glimmer of hope in this mess.

I hope this answers some of your questions and that you can convey, at least some, of my, and my colleagues, frustration and outrage over this situation, to people who have the power to do something about this.

Best,

Ron Baiman

[tommcintyre]

1. "convey...some, of my...colleagues, frustration and outrage over this..."

Thanks for posting this eomer, it DOES indeed deserve it's own thread (I was about to do the same).

I'm going to take the time to reread it, and try to convey WHY they feel so "outraged".

What I will say now is, we DO need to have the "...people who have the power...do something about this". (As Ron says)

[TruthIsAll]

2. I look forward to reading this tonight. Ron Baiman deserves our thanks

for all he has done and continues to do.
He is a patriot in the true sense of the word.

Pure scientific analysis.
Pure applied mathematics.
Pure TRUTH.

[tommcintyre]

7. A "digest" of Ron's post

<For much of us here at DU, the mathematics (though obviously necessary) can be a barrier to our understanding of what is happening. So I have "digested" Ron's excellent post in an attempt to make his information more accessible to all. (Please refer to the appropriate parts of his post above for more details.)

Thanks again for ALL your hard work at USCV. I (and many others here) are committed to getting your information before the public.)>
-----------------------------------------

"c) Liddle may have unknowingly reinvented the “wheel”...all these terms are based on the derivations developed in the April 12 USCV report..." "The importance of this is not only one of correct attribution of credit..., but also to point out that USCV and Liddle have been looking at the same variables...derived in a mathematically equivalent fashion."

"d) <Mitofsky> seems to be mistakenly, or deliberately, trying to create the impression that the Liddle analysis is based on some new, hitherto undiscovered “artifact” or “confounding”, that resolves the debate and shows that the mean rbr hypothesis can, after all, in spite of appearances to the contrary, explain the data."

"...may partly be a result of...Liddle’s paper...she derives...in a rather complicated and convoluted manner...without reference to the equations...presented in earlier USCV reports.

"e) Liddle came to an opposite conclusion to that of USCV." <Even though her conclusion "...was based on the very same...response rates patterns..."

"f) How could she have come to such a different conclusion?" "USCV reports...values diverged markedly in implausible ways across partisan precincts." "Liddle...simulated...to get the suggestive pattern displayed by Mitofsky...By looking at an exaggerated...simulation..."

"g) ...Appendix B...shows why the “inverted u” pattern appears...also points out that mean calculations that were already done in the USCV (April 12) report show that the “inverted u”...makes the jump in WPE <rBr argument>... even more implausible..."

"h) This “mathematical nit” cannot possibly explain the dramatic asymmetry in the E-M data..." <In others words, the "holes" (fallacies) in Mitofsky's rBr claim>.

"i) This debate with Liddle went on for some weeks." <With USCV pointing out to her how unrealistic her claims were.>

"j) USCV showed that...were mathematically infeasible in high Bush precincts and highly implausible in high Kerry precincts..."

"k) Liddle attempts to address this...skirts the issue." "USCV’s first report also showed that these small response rate variations contradict the E-M hypothesis."

"l) ...USCV did some output simulations...showed that...<assumptions made to support rBr were>...highly improbable to impossible..."

"m) Liddle...published her paper...was hailed as holding the key to saving the...<rBr>...hypothesis. The fact that it was based on the very same data (and pretty much the same analysis that USCV had be doing to show the opposite) was overlooked."

"n) USCV felt pressured to respond to the Liddle paper by releasing the May 21 paper, and sending representatives to the AAPOR conference. This new USCV paper in addition to the calculations...also included an input simulation that showed that the <rBr> hypothesis could not explain the <Mitofsky> outcomes, even under the most extremely (favorable to the hypothesis) <scenario>...these simulators (one of which is in completely transparent Excel form) have been put on the USCV website so that anyone can verify these results."

"o) The Liddle and E-M hypothesis had been rejected...in at least three different ways."
"a) Analysis of “representative”...precincts showed that it was mathematically infeasible or highly implausible..."

"b) “Output simulation” shows...with a high degree of certainty...matching...response rates...requires significant unexplained changes..."

"c) “Input simulation” shows that it is impossible, under the most extreme favorable to the Hypothesis circumstances, to get E-M reported results from an E-M constant... In particular this simulation shows that the E-M reported...<supporting claims>...Are all unobtainable under the E-M hypothesis."

"p) None the less, <Mitofsky>...has embraced the Liddle analysis as providing conclusive evidence for the...<rBr>...hypothesis." "The question of how such a simple...analysis could trump the data already presented by USCV...and the more recent conclusive simulation outcomes, seems to have been lost or deliberately ignored."

"q) No solid statistical evidence...is offered to support this <rBr> hypothesis."

"r) Mitofsky claimed...to have done the regressions but not released them. This is doubly unacceptable! If they were done, the public has a right to see them! <This is Mitofsky's typical obfuscating modus operandi!>

"However, I am somewhat skeptical that they have been done, or at least done in a thorough and complete manner..."

"It seems to me that this hypothesis, stated with such certainty on p. 31 of the E-M report, could not have come from an in-depth and serious statistical analysis by some of the “best analysts” in the country!

"s) ...the kind of...<data>...that E-M has released to the public, would never pass muster as supporting evidence in any kind of serious academic journal (including one that I am an editor of). The gullibility of the media in support of the constant mean bias hypothesis without any serious evidence for it has been a travesty. The notion that “these things take time” etc. is also unacceptable. The credibility of our election system is an extremely important national issue – it should not take six months or more to provide a serious analysis (especially if some of the “nations best” analysts have been looking at it) of such an important issue. Moreover, there is no reason that private business contracts or personal confidentiality should trump critical public interest in this data. There are ways to release this data that protect confidentiality (as has already been done for Ohio). There is no sufficiently important or legitimate reason for E-M not to release the data and very good reasons, relating to a minimal sense of public responsibility and survey ethics, for E-M to immediately release the data without further delay. This is what I meant by “E-M needs to release the data”.
<snip>
"z) The...<rBr>...conjecture remains an unsupported (and largely inconsistent with the data that has been made public) hypothesis. Six months after the election, we still have no serious explanation for the large exit poll discrepancy. The shoddy (if I could borrow a term) and inadequate analysis (claiming for example that tabulations and linear correlation analysis are sufficient to support the E-M hypothesis) that has been released to the public has just deepened the uncertainty about what happened in the 2004 elections. I don’t see how this could be viewed as anything other than a national disgrace. The volunteer work of USCV, and other citizen activists who are deeply concerned about the credibility and/or integrity of our electoral system, and have refused to be satisfied with this pabulum, may, in fact, be the one glimmer of hope in this mess."

[LightningFlash]

8. Regarding this, Febble keeps sending me messages....

Apparently still trying to plead her case.

First I never said she Febble was on someone's corporate payroll, that would be Mitofsky.

Second, she has already admitted to receiving money from Mitofsky. Did I say that means you are on their corporate payroll, no, and there's a difference.

Thirdly, Febble insists the bias uniform distribution is random. This is simply not true and if you fully analyze the data, and read its distinction in sequence on http://exitpollz.org, you see it clearly happens in only those selected states and in ONE direction.

So which is it? Is it uniform bias? Is it random bias?

In each case, the qualatative difference is it needs to be divergant by over twice the mean, probably around 16-24 points in other words. If someone just went around doing screwed up exit polls everywhere, the variance is in the random range of 20 points or more. This does not happen and the divergance is barely even 3-5% in ONE direction.

The very fact that this happens in only ONE direction, after the exit poll results basically switch places, is a red flag. We saw it happen in New Hampshire, with George H. W. Bush and we saw it happen in Georgia with a Senator. It's even been demonstrated in a gubernational race.

Surely if the responder bias was based in reality, it would be historically proven and archived in hindsight. Given that this is not demonstrated, and Mitofsky has kept switching his definition of "uniform" and "bias" and whatever else can be thrown at the wall, I would say the entire piece is rendered moot.

Reluctant Responder is dead. It never had any actual grounds to begin with, and there is actual factual evidence to declare vote manipulation combined with some poll problems caused the real exit poll discrepancy. :hi:

[Name removed]

9. Deleted message

Message removed by moderator. Click here to review the message board rules.

[tommcintyre]

16. A bad case of cognitive dissonance?

http://www.ethicsscoreboard.com/rb_definitions.html
Cognitive Dissonance:

Cognitive dissonance is a psychological phenomenon first identified by Leon Festinger. It occurs when there is a discrepancy between what a person believes, knows and values, and persuasive information that calls these into question. The discrepancy causes psychological discomfort, and the mind adjusts to reduce the discrepancy. In ethics, cognitive dissonance is important in its ability to alter values, such as when an admired celebrity embraces behavior that his or her admirers deplore. Their dissonance will often result in changing their attitudes toward the behavior. Dissonance also leads to rationalizations of unethical conduct, as when the appeal and potential benefits of a large amount of money <or recognition, etc.> makes unethical actions to acquire it seem less objectionable than if they were applied to smaller amounts.

IF THE SHOE FITS - WEAR IT!

[OnTheOtherHand]

20. fact check

Thirdly, Febble insists the bias uniform distribution is random. This is simply not true and if you fully analyze the data, and read its distinction in sequence on http://exitpollz.org, you see it clearly happens in only those selected states and in ONE direction.

The largest exit poll error, by either "red shift" calculation I've seen, was in Vermont.

The January E/M evaluation reports three different measures of WPE (pp. 32-33). Depending on the one you prefer, the largest errors were in Alabama, Mississippi, or Delaware.

So, while I'm not quite sure what you are claiming, I am pretty sure it is incorrect.

Surely if the responder bias was based in reality, it would be historically proven and archived in hindsight.

Proven how? (And has fraud been proven in these cases?) Here is another statement I find unclear but almost certainly factually incorrect.

Reluctant Responder is dead.

You can believe that if you want. But -- and I offer this as an empirical observation -- an awful lot of public opinion analysts disagree with you.

[tommcintyre]

38. SIMPLE QUESTION: Do you agree that USVC should be allowed to....

present the "anti-rBr position" (exit poll discrepancy indicates election fraud) at the upcoming Carter/Baker hearing at the end of June. This is a very pertinent question to ask - both YOU and Febble - since you two primarily represent the pro-rBr view here; and the pro-rBr view was given in a C-span interview as the reason for denying the anti-rBr position "a seat at the table" at the first hearing in April. (This is primarily what generated those thousands of angry emails sent by those "6 or 7 people" as Pastor described it in his call to Brad Blog. ;) )

Febble thinks "USCV should be allowed to make their case" - how about you?

Here's the relevant parts of what I asked Febble, and her relevant responses:

http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x371175#371978
"Also, for the record, do you believe that USCV should be allowed to present their case against the rBr hypothesis at the upcoming Carter/Baker Election Reform Hearing on June 30th?

Again, a CLEAR YES or NO answer would be very much appreciated.

I'm only asking you for your opinion, so please don't tell me your not qualified to answer this etc. After-all, just about EVERYONE here has an opinion on this; and you have set yourself up (and/or, been set up) as an "opinion-maker" on the pro/anti rBr issue.

BTW (just in case you're NOT aware of this): Mitofsy's rBr "hypothesis" was given as the reason that the exit poll discrepancy evidence was NOT allowed to be presented at the first C/B hearing in April."

<her response:>
"...I certainly believe that USCV should be allowed to make their case anywhere."

"And I would certainly agree that the existence of one hypothesis should never be a reason for denying the case for an alternate to be presented. It's the whole basis of cross-examination. So yes, an unequivocal YES to the principle."

So, OTOH, what do you say: YES or NO (no quibbling please - but, of course, we would all be glad to read your explanation of your opinion - either way).

Thank you,

I await your reply.

[OnTheOtherHand]

39. Autorank pointed me to your question

(My connectivity is limited right now.)

Yes, I think USCV should be allowed to present their case, and I think other strong arguments for fraud should be heard. I don't think the exit poll arguments are the strongest, but I don't see why they should be silenced just because Mitofsky disagrees with them (or even because I disagree with them, which would depend on the argument).

[autorank]

40. Thanks for responding

Now you're in a tommccintyre = autorank sandwitch.

:thumbsup:

Contact the DNC & Give 'em Hell for Not Acting on Election Frauid

NEW LEADERS FOR A NEW DEMOCRATIC PARTY

[Febble]

23. I'm not trying to plead my case

I'm trying to state it.

And I'm not even saying the bias is random, simply that it does not correlate with the partisanship of the precinct. I have shown elsewhere that it correlates with the partisanship of the state.

And I've also shown elsewhere that there has been a significant bias in the exit polls in every year since at least 1988. That may indicate fraud in every year since 1988, but there is certainly historical precedent.

[tommcintyre]

24. "since at least 1988"? Are these three posts just a coincidence?

http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x371523#372701
"Check out the 1988 NH primary. Bush Sr. was far behind Dole in the final polls He needed to win it or he was done. He got John Sununu (Gov., computer expert) to rig it for him. That was a precursor
for all that has happened since."

http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x371523#372188
Connecticut Primary 88-- the first Bush stolen election

Edited on Tue May-24-05 11:37 AM by librechik
They've been improving their technique ever since"

So now I have three (counting you) bringing up 1988 in connection with US election fraud?

Is it just a coincidence? Someone once told me, if it seems like a coincidence (especially in politics), scrutinize it very closely.

[Febble]

26. 1988 is the earliest year

for which WPE figures are given in the E-M report.

I have no information regarding previous years, which is why I said "since at least 1988".

In every year since 1988, the WPE showed "bias" - the proportion of Democratic voters (i.e. voters who voted for the Democratic candidate) polled has been larger than the proportion of Democratic votes counted.



This plot (which I derived from applying my function to the E-M WPE tables) shows the size of the bias - the "error bars" show the 95% MoE - the fact that none of them cross zero show that the bias was "significantly" greater than zero.

So each in election represented here, either Republican voters were less likely to respond to the poll than Democrats, or more Democratic votes were switched/spoiled. The math doesn't tell you which, just that one or the other occurred.

Enjoy.

[Bill Bored]

29. Any WPE data from those years Febble?

Medians, Absolute Means, Means?

I'm curious about whether the WPEs had as much variance as 2004.

On edit:

Some of the bias in previous elections has been shown to be spoiled punch cards in Democratic (African-American) precincts.

<http://www.civilrightsproject.harvard.edu/research/electoral_reform/residual_ballot.php>

[Febble]

31. Those are from the E-M report

The bars are the mean of the state WPEs, after applying my function. The state mean WPE values are from the E-M report.

The variance is the between-state variance. There is no data on the within-state variance. The between-state variance was smaller in 1992.

I think some of the bias is likely to be spoiled punch-cards in Democratic African-American precincts - the questions is, how much? You could check states you think it is likely to affect most, and see if the WPEs are higher there.

[OnTheOtherHand]

32. just a few more numbers (and a bit of history)

On p. 34 of the January report, E/M report the following WPE figures at the precinct level:

2004 average -6.5, average absolute 14.4, standard dev. 18.2
2000 average -1.8, average absolute 11.3, standard dev. 16.8
1996 average -2.2, average absolute 9.9, standard dev. 13.3
1992 average -5.0
1988 average -2.2

Notice no variance statistics before 1996. I think 1992 was the first year that VRS(? the predecessor to VNS, which was the predecessor to NEP) did a single exit poll for all networks, and they didn't compute that statistic at the time. But the tables on pp. 32-33 report state-level WPEs back to 1988, and Febble used those (I assume) to compute the error bars in her chart.

I haven't even figured out yet: if it's true, as it seems to be, that at least CBS and ABC had separate exit polls in 1988, then which set is E/M reporting? I guess whichever set Mitofsky worked on.

There's something moderately important here for the answer to tom's question, but I don't know much about it. As far as I can tell, there was no unified national exit poll until 1992. Different networks did different things in different places and ways.

ICPSR has archived the CBS/NYT and ABC exit poll data for 1984 and 1988. I think I saw someplace that the 1984 exit polls also overstated the Democratic vote share, but I haven't checked this.

(EDIT) By the way, since we're talking about WPE -- I assume that the archived exit poll data are like the data released for 2004, in that they don't include any info about actual precinct results. So they wouldn't let us study WPE. But that's an assumption; I haven't checked the codebooks.

[Febble]

33. Yes, I used the state-level

mean WPEs, applied my bias function using the state margins and computed the between-state error.

The extent to which the bias function corrects any confound with vote-count margin depends on the extent to which the mean partisanship of the precincts reflects the mean partisanship of the state, which it may not. It reduced the variance, however. What is certainly clear is that the mean WPE has been negative (i.e. greater proportion of poll responses tallied for the Democrats than proportion of votes counted for the Democrat) in every year, and certainly signficantly so in 1988, 1992 and 2004.

And bias certainly always goes the same way.

[Amaryllis]

82. See Votescam on George the first in 1988: here

http://www.democraticunderground.com/discuss/duboard.php?az=show_topic&forum=203&topic_id=331781
... did you know that George the First won fraululently? That his campaign manager Governor Sununu's "computer engineering skills approach "genius" on the tests?" "This New Hampshire primary was perhaps the most polled primary election in American history, and in the end, the Republican voters in the state confounded the predictions of nearly every published survey of voter opinion." And guess how they did it? Computer fraud.

Votescam: http://www.constitution.org/vote/votescam__.htm

[LightningFlash]

71. Again, while you state it is not so.

That does not in fact make it so just in spite. While the bias certainly was not random, that doesn't make it at all uniform, either. And in particular since your measuring tool is not neccesarily the best measuring tool for specific bias, that does not make it accurate as well.

The U.S. government's official cencus data is far more trust-worthy once again then any, one single mathmeticians "measuring" tool for bias and partisanship.

http://www.cencus.gov/Press-Release/www/releases/archives/voting/004986.html

In whole, it also alleges a very different result. One that would not be poised by anyone on Mitofsky's payroll nor anyone with their own agendas in point. It alleges that there was systematic error so widespread in the voting, that this had to have been done on computer. Error, plus numbers=fraud in two directions. Using regression of their own data you can pinpoint just where it occurs. I advocate a group of scientists to do exactly that. Debate over.

[OnTheOtherHand]

73. census data

may be official, but that doesn't make it reliable (I'm a little surprised I have to say this on DU!!!). By the way, these data are from the CPS (Current Population Survey), which isn't what most people mean by "census data" but is conducted by the Census Bureau. The decennial census attempts to count everyone in the country, but the CPS doesn't.

Here's a Census Bureau press release about the 2000 election:

http://www.census.gov/Press-Release/www/releases/archives/voting/000505.html
or http://tinyurl.com/4k6th

It ends: "Data are from the November 2000 CPS. Statistics from surveys are subject to sampling and nonsampling error. The CPS routinely overestimates voter turnout. As discussed in greater detail in the report, the CPS' estimate of overall turnout (111 million) differs from the 'official' turnout (105.6 million votes cast), as reported by the Clerk of the House."

Hmm.

[LightningFlash]

79. I'm surprised I even have to debate this with you.

The U.S. F.B.I or some "secret" order doesn't make the Cencus results. The official, confinaggled United States Cencus Bureau makes the results. They have done so by compiling Cencus data for the U.S.A the last 15 years. They probably are more of an expert than some foreign Book Writer, an Engineer, a hundred other people who are not renowned and accredited mathmeticians. That says it all.

They are accurate and in totality, and also state under every certain term that the error rare is.30%, and I'm sure they would be the LAST people to put forward a strange hypothesis like everyone lies to the cencus takers.

Their job is to report the facts instead of cover it up, what you are proposing instead is a conspiracy that would allow "vote fraud theorists" to be duped and then dan-rathered. It's even more out of the question than thinking this election was stolen.

What it provides and clearly demonstrates, is a heavy amount of fraud and areas of sampling bias.

http://www.census.gov/Press-Release/www/releases/archives/voting/004986.html

"Note how close the given MoE is to the calculated MoE (within
0.02%)using the formula:
MoE= 1/sqrt(n), where n= population size (000)

Census MoE
U.S. Total Pct Pct MoE (000) 1/sqrt(n) Diff
.Total 125,736 58.3 100% 0.30% 377 0.28% 0.02%

.Male 5
8,455 56.3 46.49% 0.40% 234 0.41% -0.01%
.Female
67,281 60.1 53.51% 0.40% 269 0.39% 0.01%

.White alone
106,588 60.3 84.77% 0.30% 320 0.31% -0.01%

..White non-Hispanic alone
99,567 65.8 79.19% 0.30% 299 0.32% -0.02%

.Black alone
14,016 56.3 11.15% 1.10% 154 0.84% 0.26%
.Asian alone
2,768 29.8 2.20% 1.70% 47 1.90% -0.20%
.Hispanic (of any race)
7,587 28.0 6.03% 1.20% 91 1.15% 0.05%

.White alone or in combination
107,930 60.3 85.84% 0.30% 324 0.30% 0.00%
..White non-Hispanic alone or in combination
100,726 65.7 80.11% 0.30% 302 0.32% -0.02%
.Black alone or in combination
14,324 56.1 11.39% 1.10% 158 0.84% 0.26%
.Asian alone or in combination
2,980 30.7 2.37% 1.70% 51 1.83% -0.13%

Table 4a. Reported Voting and Registration of the Total
Voting-Age Population, by Sex, Race and Hispanic Origin, for
States: November 2004"

[Helga Scow Stern]

12. Much gratitude to you, tom, for breaking this down and

for finding the important fact that Mitofsky has done this before to explain away Bush family victories.

[LightningFlash]

3. This is in fact what I had suspected all along...

"Liddle unfortunately does not point out that the USCV analysis was based on these very same K and B partisan response rates patterns that she investigates in her paper."

Unfortunately is an understatement, she barely even moves off her theory that there is in fact an unprecedented precinct mean bias. In this case, the uniform precinct bias should be far more divergant. It should be completely random, and in every way it is not.

It goes from one side of the country in the other, and shows the same result in up to 4 previous elections where Mitofsky first made the idea. The very concept of the reluctant responder was used as a covert "guess" in order to hide the real truth of what happened there.....And I bet being on the corporate payroll has alot to do with that.

Rest in pieces reluctant responder, we hardly knew ye uniform invisible guess.

:hi:

[tommcintyre]

4. "guess"??? I think you're being too kind ;)

"con job" would be more like it.

Even guesses (hypothesis) are backed up with evidence (However, Mitofsky even had the temerity to present this scam as a "fact"):

http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x371726
"Mitofsky does not refer to this as a hypothesis – he simply states it as a fact. Not only that, but neither in the executive summary nor in the body of the report does he provide ANY evidence to support that contention."


:mad:

[LightningFlash]

6. Mitofsky now reminds me of John Bolton...

Repeat the lie enough times, and no matter how outrageous it is eventually the gullible masses will buy it and you will be vindicated...

Sounds like he took his lessons from Karl Rove, after the so called Voters News Service became a corporate talking head.....

[autorank]

30. Interesting point. And like Bolton, I have this question about Mitofsky

Bolton is through. He'll have a job for well earned sleazoid efforts (Florida 2000, who can forget the 'yuppie riot' that stopped the Miami count). But he's finished. Everyone now knows he's just awful as a human being and a lousy diplomat.

What happens to Mitofsky after his self inflicted immolation? He's trashing himself with bogus theories and press releases arguing that he realliy doesn't know his ass from third base. How is he going to do another poll? Who would pay him? (other than Republican partisans) and Who on earth would take the results seriously?

What a flame out on his part.

[autorank]

5. Ron, thank you for every thing you're doing on FAIR ELECTIONS

I don't know you but I know your work and I am totally aware as to its importance. I am grateful to you for everything you are doing to try and get the rest of us the right to vote and know it's counted.

autorank

[tommcintyre]

10. Yes, we're lucky we have the likes of USCV and Conyers fighting for us n/t

[understandinglife]

11. Bookmarked. Nominated. Much gratitude to Ran Baiman!

Peace.


It has happened, Mr Lucas


www.missionnotaccomplished.us - how ever long it takes, the day must come when tens of millions of caring individuals peacefully but persistently defy the dictator, deny the corporatists cash flow, and halt the evil being done in Iraq and in all the other places the Bu$h neoconster regime is destroying in the name of "America."

[Helga Scow Stern]

13. Thank you for clearing this up, once again.

Was not citing the earlier equations from USCV work deliberate on the part of Liddle, or did she simply not bother to read it?

Ever since the fraudulent election, we have been seeing the same kind of obfuscation of the facts from people who are supposed to have some credibility--Mark Blumenthal, or Mystery Pollster, for instance. Or perhaps he has changed his tune. I gave up reading him once I began to doubt his sincerity.

You lay out once again another way the lies are spread, for example,

"E-M seems to be mistakenly, or deliberately, trying to create the impression that the Liddle analysis is based on some new, hitherto undiscovered “artifact” or “confounding”, that resolves the debate and shows that the mean rbr hypothesis can, after all, in spite of appearances to the contrary, explain the data."

We are learning how we are being lied to. Honest people don't expect it.

[tommcintyre]

15. Well... Kathy Dopp said pretty much the same thing. She "filched" 'em

Recently, I brought up what Kathy said:
http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x371175

"Also, according to Kathy Dopp, Febble obfuscated, and even "stole"? ("derived") much of the work:

http://www.democraticunderground.com/discuss/duboard.php?az=show_topic&forum=203&topic_id=369091#369140
"We offered to work with you on our last paper until we could all agree, and you declined. You rushed to publish your derivations of a few new formulas independently, along with your own (in my opinion) unjustified conclusions.

I've shown how simple it is to derive the three formulas that you derived on one page in our work, in Appendix C and D, since you did not reveal how you derived them yourself, and generally obfuscated the math and the issues in your own paper. Everyone else who contributed derivations of new formulas contributed them to the group and wrote up the steps to derive them so we could all easily understand, but not you.

If anyone wants to UNDERSTAND the formulas you are using clearly, they can go to our paper, which shows how easily they were derived from all our previous work.

You also obfuscated by changing all the notation we'd began with and complicating it, by calling precincts defined by percentage of votes for Kerry or Bush, Democratic or Republican precincts, etc."

[Febble]

17. Well, I didn't

Kathy is wrong.

I can supply documentary evidence in the form of email exchanges that took place before the publication of the April 12th document.

Yes, the formulas are easy to derive. And certainly I contributed them to the group, in form of a spreadsheet emailed to the group on April 7th. The April 12th edition of the USCV actually acknowledges this, although Baiman rightly also assumes responsibility for his own version, which, although algebraically equivalent, produce a different measure, as he was interested in deriving response rates, whereas I was interested in deriving the ratio between them (and taking a log).

Ron continued to deny that the ratio was a useful measure until some time later, and may still do so. He nonetheless includes a derivation in his Appendix C to his current working paper.

How either Ron or Kathy can claim that a spreadsheet sent to USCV on April 7th can be derived from a document dated April 12th is somewhat mystifying - in fact I just emailed Ron for clarification. Apparently he is not in fact claiming I "filched" the formula but derived it from his "K and B". Seeing as what "K and B" are, in Ron's notation, are the response rates for Kerry and Bush, it is not surprising that these variables appear in my formula seeing as what I was trying to derive was the ratio between them.

To my knowledge, Ron does not hold copyright in differential response rates. I thought Mitofsky was supposed to have invented those.

[tommcintyre]

19. If I recall correctly, you said Ron would vouch for you...

when this came up before. Well... it doesn't look like he will, does it? ;)

http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x372464#372698
Lizzie, I really MUST sincerely ask you - don't you think the post above about cognitive dissonance describes a lot of what's going on with you?

We are certainly all susceptible to this; but I sincerely think you are strongly afflicted with it regarding this subject. I have a feeling, overall, you're a "nice person"; and something has really gone "haywire" with you regarding this subject.

I'm really serious about this. I think you have the potential to legitimately help; but I also think this has become a serious obstacle for you.

If you want me to lay out the specific reasons I think this is so, let me know.

[OnTheOtherHand]

21. Rashomon, maybe, not cognitive dissonance

As far as I can tell, Kathy and Ron sincerely think (to paraphrase an old joke) that Febble's thinking is right and original, but that the part that is right isn't original, and the part that is original isn't right (or is at most a "mathematical nit").

Febble and I sincerely think that Kathy and Ron haven't wrapped their heads around a large part of Febble's analysis that is original, right, and important.

Febble is right to be indignant at the suggestion that she somehow -- well, I don't want to ramp up the rhetoric any more here. But I think some have suggested that she somehow stole her index, for which she reported results in a DailyKos diary on April 6, because it can be expressed in terms of K and B, which appear in the USCV paper of April 12.

It's hard to understand what this complaint could really mean. It makes about as much sense as saying that Einstein stole E=mc^2 because other physicists had discussed E, m, and c. (No, I don't think Febble's index is quite on that level, but I thought the analogy might help to clarify the issue.)

The idea of K and B extends back at least as far as E/M's famous 56%-50% sentence in January. K and B appear in the April 12 paper, but as far as I can see, neither K/B nor ln(K/B) appear there, presumably because Ron and others weren't convinced that they mattered. (For the record, as of April 12, I wasn't convinced either -- in fact, I was paying very little attention.)

Ron can continue to deny the relevance of ln(K/B) if he wants, but I don't see how he can point to the April 12 paper to deny its originality.

(EDIT TO ADD THE FOLLOWING) I think Febble is also entitled to be angry at the suggestion (if I haven't misconstrued it) that she has altered her views to follow a paycheck. I don't think anyone has challenged the factual accuracy of her assertion that she started her work for Mitofsky (you can check the quotation, I'm working from memory) about a week before the AAPOR convention. That means May. I think the public record is clear that Febble was forming her views on this issue throughout April. The portion of the private record available to me also supports this view.

[RonB]

25. Partisan Response Rates

Dear Feeble,

The Partisan response rate formulas for K and B were derived in USCV's second report back in March. I had nothing to do with these derivations. My point was that instead of just using these formulas (and perhaps their derivation with a reference to that USCV report to allow your paper to "stand alone") you derived LN(K/B) from "scratch" with entirely different symbols (and in my view much less clearly) as though the USCV work didn't exist or was not related to what you were doing. This contributed to the misleading idea that you had discovered a new "confound" or "artifact", when in fact you were simply analyzing K and B in a different way. USCV after all had looked and K-B and K/B in the March report. This is important as E-M presented your work as a breakthrough that proves the rBr hypothesis while ignoring the fact the USCV has been using the same variables (K and B) to disprove this hypothesis.

Best,

Ron

[Febble]

27. Thanks, Ron

Yes, indeed, I derived ln(K/B) from scratch.

However, the USCV certainly related to what I was doing - as I was interested in probing the point that the USCV report was making regarding the relationship between WPE and "precinct partisanship", and as you know, spotted that there was a problem.

Indeed you yourself sent me "ku doos!" (your own spelling?) on April 8th for "catching this point", and updated your paper on April 12th, acknowledging my contribution.

And as you know, I asked if you would like to be acknowledged in my paper. Although you understandably declined, I did acknowledge, on page 20, that the Appendix B derivations (the ones that had been updated in response to my 7th April email) are algebraically equivalent to mine.

As for the "misleading idea that you had discovered a new "confound" or "artifact", when in fact you were simply analyzing K and B in a different way" - I was certainly analyzing "K and B in a new way". Whether it is a way that it is useful or relevant I am happy to debate.

The fact that you yourself use it (and derive it in your current appendix c) suggests that it has at least some utility.

And E-M did not present my work as a "breakthrough that proves the rBr hypothesis" - they applied my work to demonstrate that bias as measured by ln(alpha), or ln(K/B) if you prefer, does not have a linear correlation with vote-count margin, whereas WPE did.

In other words, it removed the artefactual relationship and allowed the true linear relationship to be measured. If the fraud hypothesis does not depend on their being a linear correlation between bias and vote-count margin, then nothing has changed. We just have a better measure of bias.

The algebra was easy. The hard part was figuring out that it mattered.

[Melissa G]

28. Welcome to D U RonB!

Thanks for all your great work! We really appreciate it around here!
:hi:

[Febble]

14. Point by point:

a) I agree that this is a fair account of Ron’s work. I simply disagree with his conclusions.

b) Minor quibble: taking the log was not made at the suggestion of a member of the USCV list. I can do my own math. Major quibble: I did not “discover” the U curve when “simulating” possible values “based on the derivations developed in the April 12 USCV report”. I work daily with data expressed in percentages and knew this to be the case. I demonstrated it in my paper with an example taken from a 2:1 ratio. It was simply an example. I inferred that a wide range of ratios must exist in the actual data, which indeed it does. Many precincts, we now know, have ratios well beyond 2:1. This gives a “family” of U curves, which my formula, and also the one given in Ron’s current paper, will straighten. This is useful.

c) I have documentary evidence from emails sent round the USCV list that my formula preceded the April 12th paper, and indeed the April 12th paper was updated to reflect the implications of my formula. Ron knows this – and indeed acknowdges the fact in the April 12th paper, for which I thanked him. I have no idea why Ron wants to imply the reverse. We both agree that the WPE requires algebraic transform to remove its confound with vote-count margin, which is what matters.

d) Ron is entitled to his view of my algebra, as I am entitled to my view of his. However, his last point is wrong. I draw attention to his Appendix B on page 20 of my paper.

e) “Unvarying mean partisan” may or may not be one way of describing what I surmised. What I actually surmised was that there might (not that there would) be a linear correlation of zero once the WPE artefact was removed. There is indeed a linear correlation of zero. What else there may be is for others to judge.

f) This is a misrepresentation of my paper. I produced an initial U shaped curve to simulate a 2:1 ratio as a hypothetical illustration. I then simulated means response rates of 56% Kerry and 50% Bush as postulated by E-M. This also produced a U shaped curve with skewed distributions at the extremes. From this I concluded that the real data may show an artefactual slope between WPE and vote-count margin that would be eliminated using the ln(alpha) formula. Both these predictions were confirmed by the data.

g) In the absence of the real data I produced simulated data. Following my paper, Bruce O’Dell, at USCV also produced a simulation. We collaborated over this – I have no quarrel with Bruce and his simulator is excellent. Both of us were attempting to show whether or not the the mean and medians WPEs reported in the E-M report could be derived from set of data points that had a zero correlation between ln(alpha) and vote-count margin. I concluded that they could. Ron disagreed. We now know that they can. What other relationships may be present in the data remains a matter for discussion, and may indicate fraud. The absence of a zero linear correlation does not rule out fraud. I believe now, however, that major fraud would result in a non-zero correlation. Interested readers might like to try my own fraud simulator, posted here:.

http://uscountvotes.org/ucvAnalysis/US/exit-polls/simulators/liddle

PM me if you need instructions.

h) The simulations of what “must” be in the data are now largely irrelevant. We now know what the plot looks like, and that there are indeed four anomalous precincts in the “high Bush” category. These may be fraudulent. The are not the highest points however. There are many else where. They may all be fraudulent. My point was never to say that fraud did not occur, simply that it was not concentrated in high Bush precincts. If the anomalies in the plot are all fraud it was happening in all types of precinct, not just “high Bush” precincts. There are also some suspiciously low precincts – do these indicate pro Kerry fraud?

i) I simply disagree with this conclusion. Try the simulator yourselves.

j) This is all true. Equivalent equations are also given in my paper.

k) I simply dispute the math here. Ron’s conclusion depends on how much variance there was in the response rates – the more the variance, the less we can conclude. No data was given in the E-M report regarding variance. However, in his AAPOR talk, Mitofsky demonstrated variance for refusal rates was demonstrated, and it was high.

l) “Eppur si muove”

m) What people concluded from my paper is up to them. I have never said it disproved fraud, simply that it threw into doubt the inference that bias was greater in high Bush precincts than elsewhere. This is supported by the evidence. It may simply mean that fraud was everywhere.

n) A simulator of mine is now also available on the USCV website, so you can play with that one too.

o) If a model is unable to simulate real data there is something wrong with the model. It’s why we do models. What was wrong with both my model and the USCV model was that both understated the variance. If you increase the variance you can reproduce the data. However, my conclusion from my model was that this would be so. My conclusion was correct.

p) USCV presented an inference based on E-M data on categorical means. The plot presented by Mitofsky shows the data points that contributed to these means. They confirm that there were some very high bias rates in four “high Bush” precincts. However they also demonstrate that the reason the mean was higher in this category than elsewhere was not that similarly high bias rates were not occurring in every other category, but because else where there were also some very low values – “implausibly” low if you like – in the other categories. One just happened to miss the “high Bush” category boundary by a whisker.

q) I agree with Ron entirely

r) Agreed, except for the last part. I don’t think the math is right here.

s) Largely agree

t) Of course the WPE varies. Linear does not mean unvarying. The flat linear fit means that it does not vary linearly with vote-count margin. It may simply vary with electoral corruption. It may also vary with response bias.

u) Alpha will have a U if variance is greater at the extremes. Log alpha will not. Ron confirmed this himself here:

http://www.mysterypollster.com/main/2005/05/aapor_exit_poll.html#comments

v) “Drawing a straight line” is an odd way to describe an ordinary least squares fit. And again – alpha varies. A question for you, Ron: what do you suggest it varies with?

w) By “you” do you mean me, or someone else? I certainly don’t think those four precincts are only odd thing about the scatter. I think all those very high and very low points are odd. I just don’t know why you are so concerned about those four. As for the “only 40” high Bush precincts– it’s a Gaussian distribution – the extreme categories are the tails of a distribution. There is little in life that doesn’t have a Gaussian distribution. Precincts may be the exception I suppose, but you could probably check that. As for dropping data points – try counting them if you don’t believe me, but there are 40 little dots in that category, as in the E-M report. And in any case, I think we both agree that the centroid in that “high Bush” category is high (as long as you demarcate your category to the right of that low point at around the 79% mark.) The point is that there are plenty of high points at the high Kerry end and some low ones at the high-ish Bush end to flatten the regression line.

x) If you want to remove the outliers, you’d have to remove the other outliers too – including the low ones. What would be your z score cut-off for an outlier? It seems to me that the outliers are the interesting ones.

y) What non-linear test would you suggest, Ron?

z) I agree completely that multiple regression analyses should be reported.

Lizzie

[Febble]

18. Correction to response to point r)

I agree that mean alpha is probably closer to the figure Ron computes.

Sorry, didn't read that one properly.

[RonB]

35. On the New E-M/Liddle "Zero Linear Correlation rBr Hypothetical"

Lizzie,

As you know a zero linear correlation can be obtained from many different non-linear relationships.

USCV calculations from mean and median WPE's and mean overall response rates (R)(see Table 1 Appendix E, latest May report) show that Alpha (= K/B in this table) is highly variant and non-linear. It has one trough (for high Kerry districts where K/B is just about 1 - signifying no bias) and two peaks: in competitive districts and in high Bush districts where it reaches its highest levels of 1.55 to 1.58. Non of this is captured when fiting a flat line to this data. A quadratic, or even a third degree polynomial, would presumably fit this much more closely. The USCV report also shows that simulation of individual precincts with realistic variance etc. shows the same thing for almost all realistic simulations - as it must (see "footnote" below).

However, I think the more important point is that these calculations show that a "representative" precinct (with mean or median WPE and overall response rate) of large groups (from 40 to 540) of precincts of similar partisanship have very different mean Alpha's. The fact that a flat line can be draw through all these precincts is irrelevant to the fact that these statistically significant categorical groups of precincts have very different K/B levels. This means that a "universal" lower Bush voter response as expressed in a response bias (K/B or LN(K/B)) does not explain the different WPE levels and overall response levels by precinct partisanship.

The point here is that any "operational" or "behaviorally meaningful" definition of "rBr" has to imply that there not be statistically significant differences in bias (K/B) across different partisan groups of precincts. If there are such differences, than these need to be explained by some other factor(s) than simply "universal" rBr.
Again we're back to the need for a real substantive explanation.

The flat linear correlation (if this is the new - or old or whatever -definition of the E-M/Liddle "hypothetical" rBr non-hypothesis) really gives us very little, to nothing, with any explanatory significance for the exit poll discrepancy.

USCV focused on a non-varying mean K/B hypothesis as this could (if it were true) be evidence for an "rbr" hypothesis with real explanatory meaning. In contrast a "flat linear correlation rBr hypthetical" seems to be a relatively meaningless finding.

Footnote:


I haven't looked at Lizzie's latest simulator but if, at a high probability, it produces E-M outcomes, under the E-M hypothetical (of mean K=0.56 and B=0.5 across all precincts) there has to be something amiss with it.

If on the other hand she's simply adjusted K and B to differ across precincts so that their weighted average is 0.56 and 0.5 and so that there is a flat linear correlation - then yes she has simulated the parameters we now know about of the E-M data.

But then we're back to the point above. What does this really give us in the way of an explanation for the exit poll discrepancy? If this is all that E-M and Liddle are saying - it doesn't amount to much - and certainly doesn't justify the "commonly accepted" rBr hypothesis that the exit poll discrepancy is due to a general (not influenced by partisanship of precinct) Bush vote reluctance to respond to exit pollsters.

These mathematical games and dodges are not helpful to understanding the real problem at hand which is the exit poll discrepancy and what caused it.

Best,

Ron

[Febble]

36. Well Ron

I utterly agree, a flat line is an utterly meaningless finding.

A slope would have been a meaningful finding, but it wasn't there.

And if a linear slope isn't there, a cubic is unlikely to help. Quadratic? Doesn't look much like it to me.

As you say, it is utterly meaningless. It does not show us where the fraud is. We hoped it would, but it doesn't. It shows us, that if the high points are fraud, fraud bears no obvious relation to vote-count margin.

You will recall how resistant I was to the terms "uniform" or "constant" to describe my postulations regarding alpha. "Random" was my preferred term - that bias might have a "random" relationship with vote-count margin (or partisanship if you like). And what a "random" relationship gives you is a flatline correlation.

Of course it may not be actually random. It may be cunningly targetted at places where the statisticians will never find it. But it may still have a random relationship with vote-count margin.

I never said flat meant rBr - I just said it didn't mean Bsmvcc.

Now we know it's not Bsmvcc, let's find out what it is.

Lizzie

PS we might as well desert those means now - we can see fairly clearly that the reason the mean was low at the high Kerry was not because anything was accurate there, but because things were wildly inaccurate in both directions. You can't do much with means without variance, and now we know a bit more about the variance. It's time we started chasing those outliers, wherever they are on the plot (low ones too?)

PPS

Check out my comment here ("I'm thinkin"):

http://www.democraticunderground.com/discuss/duboard.php?az=show_mesg&forum=203&topic_id=372731&mesg_id=373020&page=

[autorank]

22. KICK

:kick:

[Peace Patriot]

34. Hey, Ron, thanks for the kudo for ordinary citizens and non-mathematici-

-ans!

"The volunteer work of USCV, and other citizen activists who are deeply concerned about the credibility and/or integrity of our electoral system, and have refused to be satisfied with this pabulum, may, in fact, be the one glimmer of hope in this mess." --Ron

We get crap like this Edison-Mitofsky item every day from Bush's theologians at the Environmental Destruction Agency, from Mr. Karl (Frosty the Snowman) Rove every time he comes out into sunlight, from the Bush "pod people" in Congress repeating their phrases over and over, and we got a rocketful of it on Feb. 5, 2003 at the United Nations.

We've developed a highly accurate nose for things Denmark.

I guess we can say this for the Bush Cartel--they're making Leftists smarter (and making more Leftists!).

And here's the answer to your plea that your frustration and outrage over this situation be conveyed to "the people who have the power to do something about this."

That would be us--the people--who won't be satisfied with pablum--who weren't satisfied with it on Nov. 2, and voted that idiot and his evil handlers out of our White House--and won't be satisfied with it, and won't put up with it, from Edison-Mitofsky, from the news monopolies who paid him to FALSIFY the exit poll numbers on everybody's TV screens on election night, or from the Darth Vaders they serve.

We have the power. We are the majority. We will find a way to restore "consent of the governed."

Thank you, Ron, for all your brilliant work! You have been an inspiration to me!

[kster]

37. kick.nt

[Time for change]

41. I've tried very hard to compare your statement with Febble's response

Thank you so much to both of you for all the work you've done on this. Regardless of who is right in this argument, you are both in favor of sufficient release of the actual data, so as to enable a more thorough analysis. I believe that that should settle this argument if it ever occurs.

I do have one question for both of you. I haven't seen Mitofsky's scatterplot that he presented at AAPOR. Is there evidence of a U shape (with strongy negative alpha or ln in the highly partisan precincts) to this scatterplot? If so, couldn't that explain why the slope is not significantly different from 0, and yet the simulations run by USCV show that it would take implausible response rates to conform to Mitofsky's data?

[Febble]

42. Links to the plots

are here:

http://www.dailykos.com/story/2005/5/24/213011/565

Plots are of ln(alpha) and WPE against Bush's share of the vote. There is also a refusal rate plot.

I don't think there is a U. Also, the weird stuff looks to me as though it is at the Kerry end, although this may be an artefact of the greater extreme of partisanship at that end, if sample sizes were small. Things will go wild at the extremes if you have so few Bush voters you have a chance of getting them all! And if you get none, the data point will disappear (alpha will go to infinity).

[eomer]

43. TFC, I believe you are right except that it is not a U shape.

If there were a U shape in the WPE data then it could be consistent with response rates of 50% and 56% across the board. Then we could argue whether response rates of 50% and 56% were plausible or implausible.

But the shape in the WPE data is something other than a U. The fact that it is not a U is the reason that it would take implausible (according to USCV) response rates to reproduce it.

Look at the graph labeled "Simulated Response Bias vs. Actual E-M Data" on page 7 of the latest USCV paper (http://uscountvotes.org/ucvAnalysis/US/exit-polls/USCV_exit_poll_simulations.pdf).

The blue U shape is the shape that would be produced by 50% and 56% response rates. The orange line is the shape of the actual WPE data.

In order to reproduce the orange line (assuming that response rates are the only source of bias) you would need the response rates shown in the graph labeled "Mean Kerry and Bush Response Rates" on page 10. These are the response rates that USCV says are implausible.

The underlying idea of your post is correct. There is a shape in the WPE data that makes the response rate explanation implausible. And this shape does give a slope of zero when it is transformed into ln(alpha) and then fitted with a regression line.

So the idea was right, it's just a different shape, not a U shape.

[Febble]

44. The blue line only has that slope

if you stick to those precise category boundaries, which are arbitrary, and not even quintiles. Move the boundary down by 1 percentage point and it drops to the level of the other categories because you include a low data point. Move the boundary up a bit and drops again, because you miss four high data points.

All the graph shows is that the mean in that particular high Bush category was high - and it is high because there are some high data points (four). There are also no real low ones.

But it doesn't mean that there are not equally high data points in lots of other parts of the plot - there are. But they didn't pull the mean up in their categories, because there happened to be low ones too. You'd expect that in higher N groups.

So the claim that there was more vote corruption in high Bush precincts still seems to me to be unsupported.

The next question therefore is - what about whole sale vote shift? But that should produce an actual across-the-board-slope, which again, isn't there. (Try my simulation - to produce the red shift with randomly distributed fraud you have to induce a slope). In fact, the plot is consistent with randomly distributed bias.

Which would be consistent with differential non-response.

It would also be consistent with more prevalent fraud in high KERRY precincts.

Which is defnitely interesting. But it is rather different to the original USCV hypothesis.

[Time for change]

45. Aren't we mixing up ln(alpha) with WPE?

The scatter plot that I was referring to in my post # 41 was a plot of Febble's "bias index" ln(alpha) vs. vote count margin. Febble, you pointed out that there does not appear to be a U shape to that, and I agree.

The blue line that eomer is referring to, I believe, is the theoretical plot of WPE vs. vote count margin if we assume there to be a uniform "bias" with a 56% Kerry response rate and a 50% Bush response rate. And this plot is very similar to what you portray in your paper.

What I'm trying to say is that I don't think that anyone is talking about the same thing here.

[Febble]

46. OK I could be confused

I thought that eomer was talking about the graph in the USCV working paper in which they fit five means to the five means in the E-M report using a combination of "vote shift" (which I don't quite understand) and response bias..

But I may have been referring to a version with different page numbers.

That's the one I find I can fit moderately well with a 56%:50% response rate, as long as I raise the variance, without any vote shift. But I'm not quite sure what they mean by vote shift.

And yes, the scatter plot you refer to does not appear to me to have a U.

In my simulation on the USCV site, vote shift is modelled by "corrupting" a proportion of precincts, and swithcing a proportion of votes to Bush - this shifts these precincts over towards the Bush end, and of course simultaneously raises their bias - producing a slope. That's the simulation scatterplot that does have a slope.

And is the reason why I think that if fraud accounts for most of the "red shift" it has to have happened mainly in Kerry territory, in order to flatten out the slope it would otherwise create.

[Time for change]

47. I was referring to the study that eomer linked me to

Well, I think that fraud in Kerry strongholds is a reasonable possibility. I don't know if I've mentioned this to you, but a recent study I conducted http://www.democraticunderground.com/discuss/duboard.php?az=show_topic&forum=203&topic_id=371211w of electronic "vote switching" reported to the national EIRS showed that of 87 reports of "vote switching" from Kerry to Bush, 42 of them occurred in the Kerry strongholds of southeast Florida -- Miami Dade, Broward, and Palm Beach.

[eomer]

48. Yes, you must be looking at the wrong graphs.

Check my references again because you're apparently looking at different graphs. The link I provided is to the latest (May 21) version.

The blue line I referenced is the result you get when you simulate 50% and 56% response rates. The resulting U shape in the WPE is not an artifact generated by slicing. It is the U shape you expect to get no matter how you slice it.

As far as the rest of what you said, can you be more specific about which graphs you are talking about? Hopefully you can reference the latest version because I don't have all the prior ones.

[Febble]

50. yes it is a U

but a tilted U, and if you make the variance high enough there is not a bad fit to each point (somewhat better than the overall fit of the scenario modelled in the paper, I would argue.)

And the scatter plot of course is the whole family of tilted Us, which the index function straigtens out - so instead of looking like a tilted football, it looks like a horizontal sausage.

Have a look again at the WPE scatterplot to see what I mean - the really high values are nearer the middle. And if you also eyeball each category, you can find near enough the centroid - which is indeed high for the high Bush category, but not because that category has the highest points, but because it doesn't have low ones - actually the middle categories have the highest values. And again, the high Bush category only misses out on a low one by a whisker.

Which is not to say the high Bush mean isn't high. But that the category means are a) acutely sensitive to where you draw the boundaries and b) small N categories are even more sensitive than large N categories, which is why usually people use percentile splits (with equal Ns) if they want to slice at all.

I suppose what I'm saying is that if there were error bars on those data points, the fit you get from specifying 56% and 50% and a variance of .2 wouldn't look any worse than the fit Ron and Kathy have done. And if you moved the high Bush boundary up a bit, or down a bit, you'd get very different means.

Which is why I think the scatterplot is more informative, and why I think that if fraud is going to account for the net red-shift, it has to have been more prevalent in high Kerry precincts than in high Bush precincts, the opposite of the USCV initial (but understandable - I shared it) conclusion. There are some very odd things going on in that top left hand corner (WPE_index plot).

[eomer]

51. Slicing and dicing...

I think we're mostly on the same page. With regard to moving the boundaries I can't argue but it sure would be nice if Mitofsky would release a data version of the two plots (maybe you could use your relationship with him to suggest that). Then we could move the boundaries around and see how sensitive the category means are to slight differences in the boundaries and we could slice it into equal N slices to see how that looks and probably many other interesting things.

On your point that your fit is just as good as the fit that Kathy and Ron have done - I don't find their fit very convincing either. OTOH pointed out a while back that the fit isn't that great and I can't argue. The fit they've produced isn't good enough to fully explain the data. In their defense, they seem in that part of the paper to be pursuing what some have called the "fingerprint" of fraud and are apparently focusing more on the inflection points of the shape than how well it fits in the regression sense. And further in their defense, they say they can improve the fit with small adjustments in the response rates. That said, I would like to see what adjustments in the response rates they need to make and then how good the fit is. Until then I don't find that fingerprint shape part of the paper very convincing.

So back to your simulation, if your fit is just as good as theirs then I don't think it's good enough yet to really explain the data.

I agree with your other points but would refine what you are saying about the fraud having to be more prevalent in Kerry strongholds. I agree but we still need a relatively universal effect in order to lift the whole thing above zero. So you probably need some near universal effect, which would lift the whole thing but create a slope in the WPE_Index regression line, coupled with extra effect in the Kerry strongholds in order to pull the left end of the regression line back up to near zero slope.

The bottom line of this discussion is that no way has rBr been shown to be true nor has it been shown to be any more coherent or consistent with the data than fraud.

[Febble]

52. Yes, I think we might be!

Have you seen this diary on Dkos?

It contains links to updated versions of the plots with axes marked, which should help quite a bit,if you haven't got them already.


http://www.dailykos.com/story/2005/5/24/213011/565

The think I keep banging on about is the variance. Everyone got hooked on this 56% 50% business, but the variance is where the story is, if there is one. It it HUGE. Ron calls my little 2:1 curve exaggerated (because 56%:50% is only 1.12 to 1) but actually there are loads of points far more extreme than that in both directions.

And no, I don't think my fit explains the means data particularly well, but I don't think it has to. My point was always that lots of scenarios could fit those means given enough variance (and the variance was always partly deducible from the absolute WPEs). But we don't have to infer what values make up those means anymore as we have the plot. Yes, it would be nice to have more than the plot, but the plot tells us one heck of a lot more than the means do.

It tells us a) the variance was huge b) that there were extreme data points in both directions c) that the bulk of less extreme data points is probably enough to account for the whole red shift (i.e. even if you slice off high and low outliers) and d) bias doesn't vary linearly with vote-count margin.

Not only that, but we know something about the variance in refusal rates (plot also linked from the Dkos diary), which again is collosal.

Your second last paragraph is where I am at present. And I would agree that so far differential response rates have not been shown to be very much more convincing than what I think is the remaining fraud hypothesis, of massive fraud that was considerably more prevalent in Kerry strongholds.

As for your last: I doubt if either will be ever be "shown to be true". All we can do is hypothesise patterns that are consistent with both, and, ideally, would distinguish between the two. Not easy.

But I agree that there has been no slam dunk, as yet, as far as I can see, for either team.

[eomer]

53. Yes, I saw that diary and bookmarked it...

when you posted a link previously. Thanks.

[eomer]

49. Response to TFC #45

TFC, the way I understand it is that you won't get a U shape in a plot of bias index if the data you are plotting is based on a uniform bias. It is in a plot of WPE based on uniform bias that you will get a U shape. Then if you apply Febble's function to that WPE to turn it into bias index the result should be that the U is flattened out into a straight line.

In other words, a plot of the bias index that is based on a uniform bias should reflect a uniform bias (a straight line).

To be clear about what I'm saying, it is the fact that the actual WPE data does not have a U shape that shows it is not produced by uniform bias.

In order to reproduce the shape of the orange line that I referenced (that looks almost like a Z lying on its side) by way of response rates, you need response rates that seem to be implausible. That orange line is the shape of the actual WPE data when you slice it into five bands.

I guess that Febble is saying she can reproduce that same orange line by increasing the variance but I'm not sure because she wasn't specific about which line on which graph she can reproduce. If it is the same orange line that I'm talking about then I question whether increasing the variance can reproduce it. The only way I can see you could get there is by an occasional accidental match produced by random variation. If that's what Febble means when she says she can reproduce it then the question is - how often does this accidental match occur and based on what variance?

[Time for change]

54. Thank you for the clarification eomer

Actually, when I asked the question about the U shape I was referring to Mitofsky's scatterplot (which I hadn't seen until Febble linked me to it), recognizing that there wasn't supposed to be a U shape to it, but thinking that if there was it could explain why the regression line was not statistically different from zero, even in the presence of high bias in the Bush strongholds. It was just a shot in the dark really, and I asked the question partly because I couldn't think of a better question, and it looked like this thread was dying away, and I thought it should be resurrected.

Since then I've done a lot of thinking about this and have also had some private e-mail discussions with Febble about her paper, which helped much in improving my understanding of this whole thing.

The main conclusion I came to from all this was that it is quite misleading to imply (as I believe the MSM has done) from the fact that Mitofsky's scatterplot does not have a statistically significant positive slope that therefore his scatterplot supports rBr.

I have detailed my reasoning on this in a new thread: http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x373817.
The reasons that I started a new thread on this were 1) I feel that this is a critical issue which deserves a separate discussion on a new thread, and 2) I believe that this thread, beginning with Ron Baiman's post, is too technical for the great majority of DUers to follow.

[RonB]

55. E-M's "Hypothetical" rBr and Zero Correlation Do Not Explain Exit Polls

Dear “Time for Change”,

I tend to agree with you that part of the explanation could be the non-linear variation of K/M across partisan precincts. As can be seen from the following table LN(K/B) (Febble’s index), does vary significantly across partisan precincts. This implies that a flat, uniform non-varying LN(K/B) cannot adequately produce the shape of mean and median WPE and overall response values reported by E-M. In particular, the mathematical asymmetry generated by comparing a ratio (K/B) to an absolute (WPE) is not plausibly adequate to explain the big jump in WPE in high Bush precincts, even after outliers are removed (the median calculations). As other respondents to your post have noted, a constant K/B bias produces a slightly asymmetric inverted WPE "U" pattern but not to the degree, and with the strong increase in mean and median (negative) WPE in high Bush districts, shown in the E-M report.

I’m copying below a post that I recently sent out on the USCV list. I apologize to DU readers if it is a bit technical for this forum.

Dear Colleagues,

Attached (and copied below) are tables which analyze:

a) The newly released (at AAPOR) E-M finding that here is a zero linear correlation between LN(K/B) and precinct partisanship, and

b) The (correct) assertion made by Mitofsky (in a private communication) that the K=.56 and B=.50 is stated as a "hypothetical" on p.31 of the E-M report and not a "hypothesis". Mitfosky claims that E-M never claimed that K/B was "unvarying", just that hypothetical values of K=0.56 and B=0.50 could explain all of the exit poll discrepancy.

My claims are that:

a) The zero linear correlation between LN(K/B) and precinct partisanship is a mathematical finding of little to no operational, or practical, value in terms of explaining the exit poll discrepancy. In particular, it does not, as E-M suggested at AAPOR, prove that K/B is randomly distributed across precinct categories. Table's 2 and 4 show that, at the 95% two tail confidence level, it is highly unlikely (the analysis is done off of aggregate data as we don't have the real data so there is some approximation involved - but approximation that is highly unlikely to change the main conclusions) that the variances in K and B levels between quintile precinct categories are a result of random error. In particular, the B level for competitive precincts differs significantly from all other B levels, and the K level for High Bush precincts differs significantly from all other K levels. This latter finding would seem to provide support for the Bsvcc hypothesis.

The general point is that a zero linear correlation of all the data with precinct partisanship does not prove that statistically significant subsets are not affected by precinct partisanship, perhaps in a non-linear fashion. A "mean uniform" bias hypothesis thus cannot explain the discrepancy. If bias is not pervasive and uniform on average, than the differing levels of "bias" in different categories of precincts have to be separately explained. The rBr hypothesis is thus an inadequate explanation of the discrepancy, regardless of the zero linear correlation. We are still left with no serious (multifactor statistical) explanation from E-M and no access to data that would allow other independent parties to do this analysis.

Tables 1 and 2 perform the analysis for means, and Tables 3 and 4 do the same with medians. Analysis with medians should pretty much eliminate the undue influence of extreme outliers. The far right hand side columns of Tables 1 and 3 also look at whether an unvarying bias of K=.56 and B=.5 could explain the K and B differences and find that they cannot.

Tables 5 and 6 show that random deviations of weighted average K and B, of whatever value, can also not fully explain the K and B differences across precinct categories.

b) Mitofsky stated at AAPOR that he had done the (multifactor) regressions but not released them. It is unacceptable that this analysis (which has absolutely no bearing on respondent confidentiality) has not been released. Moreover, presumably this analysis was at the basis of the "hypothetical" statement that K=.56 and B=.5 could explain the discrepancy. USCV (and almost everyone else) interpreted this as meaning that a mean uniform rBr of this magnitude was "the explanation" of the exit poll discrepancy. It is possible that, based on this analysis, E-M believe they have found an explanation for the discrepancy that did not rise to the level of critical national importance (like vote fraud) but none the less (they believe - perhaps related to their commercial interests) is too embarrassing or controversial to release publicly. If true this could, (perhaps!), be justified as a necessary compromise between public responsibility and commercial interest (why are exit polls not paid for by a non-partisan public funding for the explicit purpose of verifying election accuracy?).

However, as Tables 7 and 8 below show, it is simply not possible, even under the most extreme (bias minimizing assumptions) that weighted average (not unvarying) values of K=.56 and B=.5 could explain the exit poll discrepancy. These tables use the "theorem" proved in Appendix B of the last two USCV reports that the more competitive a precinct, the greater the WPE a given K/B level (or K-B gap) generates. The precincts in Tables 7 and 8 are thus chosen to be as competitive as possible within their respective quintiles.

My conclusion is that whatever analysis has been done is most likely inadequate and incomplete, or simply erroneous. Therefore, though E-M may believe that they have found a real explanation (not a zero correlation or "hypothetical" rBr that really don't explain anything) and may sincerely believe that withholding it is in the public interest (as well as perhaps their commercial interest or perhaps just the later but not to the detriment of the former) - as the "hypothetical" conclusion of their analysis appears inconsistent with other data that they have provided, I don't believe that we the public can have confidence in the actions, and inactions, that have been taken. This is all the more reason to release the data (in an appropriate form as has, at least partially, been done for one party for Ohio) and allow independent analysts, like USCV, to reach their own conclusions regarding an exit poll explanation, or lack thereof, of the exit poll discrepancy.


Best,

Ron

P.S - I apologize for formating problems which make many of these tables very difficult (nearly impossible) to read. The smaller ones: Tables 2,4,7,and 8 are not too badly mangled!
Table 1: Mean Based Calculations: Probability of K and B as is and in Relation to K=.56 and B=.5

Precinct Vote
Bush Kerry R Mean WPE Kk/R Bb/R SD for Proportion Sample Size Prob Kk/R Prob Bb/R Sample Size B SD B Hyp B Hyp K Sample Size K SD K Prob(K)=.56 Prob(B)=.5 K/B LN(K/B)
0.1 0.9 0.53 0.3% 89.9% 10.2% 0.031623 90 48.1% 48.1% 90 53.8% 16.8% 50.0% 56.0% 90 52.9% 1.9% 4.9% 41.0% 0.984 -0.0166
0.3 0.7 0.55 -5.9% 73.0% 27.1% 0.035675 165 20.4% 20.4% 165 49.6% 6.5% 50.0% 56.0% 165 57.3% 2.8% 68.1% 47.5% 1.156 0.1448
0.5 0.5 0.52 -8.5% 54.3% 45.8% 0.021517 540 2.4% 2.4% 540 47.6% 2.2% 50.0% 56.0% 540 56.4% 2.2% 57.4% 14.0% 1.186 0.1704
0.7 0.3 0.55 -6.1% 33.1% 67.0% 0.022495 415 8.8% 8.8% 415 52.6% 1.8% 50.0% 56.0% 415 60.6% 4.1% 13.3% 7.0% 1.152 0.1414
0.9 0.1 0.56 -10.0% 15.0% 85.0% 0.047434 40 14.6% 14.6% 40 52.9% 3.0% 50.0% 56.0% 40 84.0% 26.6% 14.6% 16.4% 1.588 0.4626


Table 2: Mean Based Calculations:

95% Confidence Interval for B 95% Confidence Interval for K
Covers all B? -1.96 x SD' B +1.96 x SD' Covers all K? -1.96 x SD' K +1.96 x SD'
Yes 20.9% 53.8% 86.6% No 49.3% 52.9% 56.6%
Yes 36.8% 49.6% 62.4% No 51.8% 57.3% 62.8%
No 43.2% 47.6% 52.0% No 52.0% 56.4% 60.8%
Yes 49.1% 52.6% 56.1% No 52.5% 60.6% 68.7%
Yes 47.1% 52.9% 58.7% Yes 31.9% 84.0% 136.1%

Table 3: Median Based Calculations: Probability of K and B as is and in Relation to K=.56 and B=.5

Precinct Vote
Bush Kerry R Median WPE Kk/R Bb/R SD for Proportion Sample Size Prob Kk/R Prob Bb/R Sample Size B SD B Hyp B Hyp K Sample Size K SD K Prob(K)=.56 Prob(B)=.5 K/B LN(K/B)
0.1 0.9 0.53 -0.4% 90.2% 9.8% 0.031623 90 47.5% 47.5% 90 51.9% 16.8% 50.0% 56.0% 90 53.1% 1.9% 6.1% 45.4% 1.023 0.0224
0.3 0.7 0.55 -5.5% 72.8% 27.3% 0.035675 165 22.0% 22.0% 165 50.0% 6.5% 50.0% 56.0% 165 57.2% 2.8% 66.1% 49.7% 1.144 0.1347
0.5 0.5 0.52 -8.3% 54.2% 45.9% 0.021517 540 2.7% 2.7% 540 47.7% 2.2% 50.0% 56.0% 540 56.3% 2.2% 55.6% 15.0% 1.181 0.1664
0.7 0.3 0.55 -6.1% 33.1% 67.0% 0.022495 415 8.8% 8.8% 415 52.6% 1.8% 50.0% 56.0% 415 60.6% 4.1% 13.3% 7.0% 1.152 0.1414
0.9 0.1 0.56 -5.8% 12.9% 87.1% 0.047434 40 27.0% 27.0% 40 54.2% 3.0% 50.0% 56.0% 40 72.2% 26.6% 27.0% 7.8% 1.333 0.2874

Table 4: Median Based Calculations:

95% Confidence Interval for B 95% Confidence Interval for K
Covers all B? -1.96 x SD' B +1.96 x SD' Covers all K? -1.96 x SD' K +1.96 x SD'
Yes 19.1% 51.9% 84.8% No 49.5% 53.1% 56.8%
Yes 37.1% 50.0% 62.8% No 51.7% 57.2% 62.7%
No 43.3% 47.7% 52.1% No 51.9% 56.3% 60.7%
Yes 49.1% 52.6% 56.1% No 52.5% 60.6% 68.7%
Yes 48.4% 54.2% 60.0% Yes 20.2% 72.2% 124.3%

Table 5: Mean Based Calculations: Probability of K and B in Relation to Weighted Average of K and B

Precinct Vote
Bush Kerry R Mean WPE Kk/R Bb/R SD for Proportion Sample Size B Hyp B Hyp K Sample Size K SD K SD B Prob(K)=.586 Prob(B)=.501
0.1 0.9 0.53 0.3% 89.9% 10.2% 0.031623 90 53.8% 50.1% 58.6% 90 52.9% 1.9% 16.8% 0.1% 41.3%
0.3 0.7 0.55 -5.9% 73.0% 27.1% 0.035675 165 49.6% 50.1% 58.6% 165 57.3% 2.8% 6.5% 33.0% 46.7%
0.5 0.5 0.52 -8.5% 54.3% 45.8% 0.021517 540 47.6% 50.1% 58.6% 540 56.4% 2.2% 2.2% 17.0% 12.7%
0.7 0.3 0.55 -6.1% 33.1% 67.0% 0.022495 415 52.6% 50.1% 58.6% 415 60.6% 4.1% 1.8% 31.1% 8.1%
0.9 0.1 0.56 -10.0% 15.0% 85.0% 0.047434 40 52.9% 50.1% 58.6% 40 84.0% 26.6% 3.0% 16.9% 17.5%
1250 626.63 1250 731.92
Weighted Avg 50.1% Weighted Avg 58.6%

Table 6: Median Based Calculations: Probability of K and B in Relation to Weighted Average of K and B

Precinct Vote
Bush Kerry R Median WPE Kk/R Bb/R SD for Proportion Sample Size B Hyp B Hyp K K SD K SD B Prob(K)=.581 Prob(B)=.501
0.1 0.9 0.53 -0.4% 90.2% 9.8% 0.031623 90 51.9% 50.1% 58.1% 90 53.1% 1.9% 16.8% 0.4% 45.7%
0.3 0.7 0.55 -5.5% 72.8% 27.3% 0.035675 165 50.0% 50.1% 58.1% 165 57.2% 2.8% 6.5% 36.5% 48.9%
0.5 0.5 0.52 -8.3% 54.2% 45.9% 0.021517 540 47.7% 50.1% 58.1% 540 56.3% 2.2% 2.2% 20.9% 13.7%
0.7 0.3 0.55 -6.1% 33.1% 67.0% 0.022495 415 52.6% 50.1% 58.1% 415 60.6% 4.1% 1.8% 27.5% 8.1%
0.9 0.1 0.56 -5.8% 12.9% 87.1% 0.047434 40 54.2% 50.1% 58.1% 40 72.2% 26.6% 3.0% 29.8% 8.4%
1250 626.65 1250 726.58
Weighted Avg 50.1% Weighted Avg 58.1%

Table 7: Mean Based Calculations: Minimal Weighted Average B and K

Precinct Vote
Bush Kerry R Mean WPE Sample Size B Sample Size K
0.19 0.81 0.53 0.3% 90 53.4% 90 52.9%
0.39 0.61 0.55 -5.9% 165 50.8% 165 57.7%
0.5 0.5 0.52 -8.5% 540 47.6% 540 56.4%
0.61 0.39 0.55 -6.1% 415 52.3% 415 59.3%
0.81 0.19 0.56 -10.0% 40 52.5% 40 70.7%
1250 626.75 1250 721.81
Weighted Avg 50.1% Weighted Avg 57.7%

Table 8: Median Based Calculations: Minimal Weighted Average B and K

Precinct Vote
Bush Kerry R Median WPE Sample Size B Sample Size K
0.19 0.81 0.53 -0.4% 90 52.4% 90 53.1%
0.39 0.61 0.55 -5.5% 165 51.1% 165 57.5%
0.5 0.5 0.52 -8.3% 540 47.7% 540 56.3%
0.61 0.39 0.55 -6.1% 415 52.3% 415 59.3%
0.81 0.19 0.56 -5.8% 40 54.0% 40 64.5%
1250 627.48 1250 718.68
Weighted Avg 50.2% Weighted Avg 57.5%

[Time for change]

56. Thoughts on significance of non-significant slope

Thank you very much Ron for the extensive explanations. I have to say that, as someone who is not primarily a statistician (though I do frequently use statistics in my work as an epidemiologist) I have a good deal of difficulty in following much of your language. Nevertheless, I have read some of USCV's papers, including your response to Mitofsky's presentation, I have spent a good deal of time reading your discussion on this post (and Febble's response), I have discussed these issues in some depth with Febble (and read her paper)and others in this forum, and I have been thinking a great deal about these issues lately.

Consequently, I have developed some ideas on this subject which I would like to share with you (and others), as not only would I be interested in your response to them, but perhaps you might find them useful (Since I consider the 2004 election to be the most important issue in the world right now, I jump at any opportunity to be of some use to the discussion). It's not absolutely clear to me whether USCV has already considered these ideas, or if I'm saying some of the same things in different words, but here they are:

1) Here are some ideas that I recently posted explaining why I believe that Mitofski's "non-significant" slope not only doesn't support rBr but actually argues against it:
http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x373817
I would be interested in your response to my first two points.

2) Later on in the same thread, in a discussion with Febble, I make the broader point of why I don't believe that a significantly positive slope is a good argument against fraud:
http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x373817#374019

3) When I bring up your simulations to Febble she says to me that she thinks that they're wrong. It is not totally clear to me why she thinks that (maybe I haven't pressed her hard enough on that point), but I think that it has something to do with the fact that she thinks the the assumptions on which your simulations are based are wrong.

But here's what I don't get about that: It seems to me that assumptions or even simulations aren't even necessary. Considering the mean WPE in the Bush strongholds (>=80%) of -10.0, and the median of -5.8, that means that the 20 precincts with the most negative WPE in the Bush stronghold would have at the very least (least negative that is) a mean WPE of -14.2. Making the conservative estimate of a mean WPE of -14.2 and a Bush vote count (according to the polls) of 80%, and assuming a total sampling rate of 56%, one could show through some algebraic calculations that the Kerry sampling rate would have to be 87% (for this group of 20 precincts as a whole), the Bush rate would have to be 51%, and alpha would be 1.67. No assumptions necessary, except that the data Mitofsky provided are correct.

4) I've looked very hard at Mitofski's plot of ln(alpha) vs. Bush vote count, and it doesn't look non-significant to me. Precincts with ln(alpha)>.5 and Bush vote count>=80% look very numerous to me, whereas the number of precincts with ln(alpha)>.5 and Bush vote count<=20% looks very sparse. Are we sure this line is not significantly positive?

[OnTheOtherHand]

59. on what response rates "have to be"

I think this line of argument (e.g. in your point 3) is appealing, but basically incorrect.

Consider, for a moment, a single precinct in which there is no fraud, which votes 80% Bush and 20% Kerry. Suppose the exit poll broke 70% Bush and 30% Kerry in that precinct. So we could say that obviously the Kerry response rate is at least 50% higher than the Bush response rate, in fact more than that -- so maybe the Bush response rate is 50% and the Kerry response rate is near 80%, which seems bizarre.

But it is actually entirely unnecessary. It is perfectly possible, through sampling error alone, to approach (say) 40 people, have half of them respond, and have (say) 14 say they voted for Bush and 6 say they voted for Kerry. In fact, in this example, we would expect 6+ Kerry voters in the sample about 20% of the time (or more, if we suppose that Kerry voters participate at slightly higher rates). It's true that 6 is 50% more than the 4 Kerry voters out of 20 respondents that we expect, but it isn't true that the response rate has to be 50+% higher.

I think this is the basic problem with most of Ron Baiman's arguments about response rates.

[Time for change]

60. It seems to me that you're basing your argument on small numbers

What I'm saying is that if you take the 20 Bush stronghold precincts with the most negative WPEs, it can be shown without making any assumptions (except that there was no fraud) that the Kerry sampling rate was 87% and the Bush sampling rate was 51% (alpha=1.67) for all 20 of them combined -- and that's a conservative estimate. That's a huge difference, and I would suspect that the combined numbers for all 20 precincts is large enough that sampling error wouldn't come close to explaining it. And furthermore, it seems rather odd to me that the sampling rate of Kerry voters in Bush precincts would be so high.

[Febble]

61. The numbers are small

in two senses.

The sample sizes were small (the E-M report tells as the average was (79) and there is a cluster of four precincts in that high Bush category in which the ln(alpha) is over 1, giving an alpha of probably around 3 (Kerry voter sampling rate 3 times higher than Bush sampling rate). Something very odd may have gone on in those four precincts.

But there are around 12 precincts at about that level or higher elsewhere on the plot. So my problem is not with the conclusion that those four were odd, but with the generalizability of the conclusion - that "Bush strongholds had more vote-count corruption". Well, it's one of my problems. Here is another: there are almost as many precincts with an alpha of .33 - i.e. (Bush voting rate three times higher than Kerry voting rate). So large differentials are certainly not unprecedented elsewhere, and are certainly not confined to Kerry-sampled-more-than-Bush scenarios.

But my biggest problem is that it isn't these very extreme points that are lifting the mean. The really striking thing about the plot is that if you draw a line through either the .5 or even the -.5 mark you hit a lot of precincts. And .5 is an alpha of 1.6. An alpha of 2 (the one Ron thinks is exaggerated) gives a ln(alpha) of 0.7 - and again, that line hits a lot of data points, mostly in the middle (because that is where most of the data points are).

And even below zero, there are a large number of data points as far below zero as the mean is above zero. So even the mean alpha of around 1.12 is far from unprecedented in the opposite direction.

So what all this is saying to me is that extreme alphas are not, in themselves "implausible" - they happen the whole time, in both directions. So unless you are willing to say that large ln(alphas) in both directions are evidence of fraud - and you'd have to implicate Democrats here - then all we are talking about is a numerical superiority of positive rather than negative "implausible" alphas. Which is odd, and needs explaining.

But to say that their very magnitude is "implausible" and therefore indicates fraud seems to fly in the face of the evidence.

More sensible, to me, would be the conclusion that alpha just is very variable, and when something (fraud or underlying differential compliance rates)pushes them upwards, it pushes them all upwards. And if fraud was pulling them upwards, I think I'd expect a positively skewed distribution - whereas in fact I think all we have is a normal distribution with a shifted mean.

Unless, of course, there is substantial pro-Kerry fraud. Which I am not suggesting.

I suppose, to use an epidemiological metaphor - if you were worried about obesity in a population, and you observed that mean body weight was increaing - you might wonder whether it was being leveraged upwards by a proportion of very obese people, or whether everyone was just a little more overweight than they used to be. If the former, you might think that some particular agent was causing obesity in a proportion of people. If the latter you might think that some more general process was going on (everyone eating more chips).

To me, it looks like the latter. There are still very thin people and very fat people, but everyone's a little fatter than they were.

It is only in that high Bush category that the mean is being leveraged by four obese precincts, and even then only because a skinny precinct doesn't quite make it into the category. But would that change your conclusion, as an epidemiologist? If there were many similar clusters it might, but I'm not sure that one cluster makes an epidemic!

Which is why I tend to the view that whatever is pushing the mean upwards is extremely pervasive. Which won't rule out fraud for some people, but is starting to do so for me (massive fraud at least). Especially as I think fraud on that scale would give you a positive slope.

[OnTheOtherHand]

63. no, think about this again

I think you are making a good point, but you aren't entirely engaging my point.

In my example for a single precinct with 20% Kerry voters and a 14-6 exit poll response, the calculated alpha would be something like 1.71 -- yet the actual ratio of response rates in that precinct is 1.00 (or could have been pretty much anything I stipulated it to be). Since alpha applied post hoc doesn't actually tell us the ratio of response rates, it won't be possible to say anything post hoc about response rates without making some assumptions!!

(I think I missed this development in your thinking, but I don't think that talking about "sampling rates" instead of "response rates" solves the underlying problem here. You may need to point me back to another discussion.)

Now, I'm not claiming that individual-level sampling error alone can account for the Bush strongholds. Obviously there is huge variation across the precincts. But if alpha can fairly routinely be off by 70% or more in individual precincts ("more" is very possible -- my 14-6 was hardly extreme), then I'm quizzical about what our calculated response rates really mean.

Here's something I think I can suggest to an epidemiologist: Check out Bruce O'Dell's working paper at http://tinyurl.com/8mtfe (it's hosted on his corporate website, I'm just sick of watching DU truncate URLs!), and download his simulator from the link on p. 10. Go to the high-Bush precincts (B83 tab #1 -- use tab #1 because, as you can verify, it comes closer than tab #2 to reproducing E/M's reported mean abs WPE of 12.4%) and calculate ln alpha, a standard error for ln alpha, and a confidence interval of your choice. Tell me what you think about your results.

[Time for change]

64. Sorry I'm not entirely engaging your point

This post is in response to both you and Febble, since I believe that there is a lot of overlap.

The reason that I'm not entirely engaging your point is that I don't entirely understand your point. I can't read your post without having it seem to me that your main problem is with perceived small sample size -- or if there is another problem I don't understand what you're saying. Sure, in the example you give there is a 20% chance of obtaining the outlying result that you posit. But when the sample size becomes much bigger then, as you know, that problem goes away.

Febble also adds that sample size is a major problem with my idea. I don't understand that, since I'm talking about the means for 20 precincts.

Having said that, I need also to admit that the calculations that I noted in my previous post were off. Instead of a Kerry sampling rate of 87%, that should be 76%. But that assumes only an 80% Bush voter percent for the group, which is very conservative for the 80-100 category. Making a more reasonable assumption of 85%, the Kerry sampling rate rises to 82.7%.

So let me clarify:
For the 20 precincts in the Bush strongholds with the most negative WPE, assuming a Bush vote count of 85% for the group and a mean WPE of -14.2, we get a Kerry sampling rate of 82.7% and a Bush sampling rate of 51.3%. It is also important to mention that this differential sampling rate is the absolute minimum, because it assumes a mean WPE of -14.2, whereas that is the very least negative WPE that is possible from Mitofsky's data. The real WPE is certainly more negative than that, and possibly quite a bit more negative. I don't think that is plausible. If you do think it's plausible, is that because you think the numbers are small enough that sampling error could be a problem, or is there some other reason? Maybe you've already explained it above, but I just don't get it.

Maybe it's because you consider the 20 precincts to represent "cherry picking" from the 40 available precincts. Well, perhaps it is, but still I think that the differential is quite striking.

So what happens if we look at all 40 precincts? Still (assuming 85% Bush vote share) we get 74.7% Kerry voter sampling rate, vs. 52.7% Bush sampling rate (yes, 3 significant figures is probably too much). No cherry picking there. Considering that this applied to all 40 Bush stronghold precincts as a group, I think that this figure is also quite striking, though I still consider the figure for the 20 precincts more striking, notwithstanding the fact that they are a smaller and a select group. Anyhow, I just don't get the argument that this result can be discounted because of small sample size. What do you think the sample size is, anyhow?

Regarding the issue of sampling vs. response rate, I can't point you to a post. I believe that Febble and I discussed it by pm. Sampling rate is a simpler concept, because it is simply the number of responses, divided by the total attempted sample (the number of responses plus the number of refusals plus the number of misses). "Response rate" is the number of responses divided by the number "approached", so it would avoid the misses. By this definition, the numbers that Mitofsky presents would be sampling rate, rather than response rate. So I just think it gets unnecessarily complicated when you start talking about response rate (though I'm pretty sure that the great majority of the time, when someone uses that term, they are really referring to sampling rate, by the above definition.)

[TruthIsAll]

62. To: Ron, Bruce, Kathy, TFC, EOMER, FEBBLE, OTOH

Have any of you seen this?

http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x374165

Optimization has not been used to analyze rBr - until now.

The Excel SOLVER optimizer returned a NO FEASIBLE SOLUTION message when I tried to goal-seek the model to produce 51.2% for Bush (his 2-party vote). The maximum iteration count was exceeded as the model tried it's darnedest, but instead the iteration moved in Kerry's direction. There was NO response distribution which could reproduce the 51.2% goal assuming a 53% weighted average response rate. He only gets 51.2% if the average rate is 50% in conjunction with an implausible response rate in Kerry strongholds. See the thread for details.

The results confirm the USCV simulation, but go one step further: rBr is not only implausible, it's also unfeasible. The analysis iterates nicely to that conclusion.

If any of you have any problems with the analysis, feel free to make suggestions. I would gladly re-run the optimizations under any criteria, constraints or scenarios you would like.

I appreciate your comments.

[Time for change]

65. Thank you for the information TIA

I think that I should refrain responding to this, because I think that among the people to whom you addressed this post, most or all of them understand the simulation and modeling process that you refer to better than I do.

[autorank]

70. So if "most of all of them understand the simulation process"

they why don't they go to TIA's thread? You still don't find this curious? Quelle domamge...

[OnTheOtherHand]

72. actually I was there

A model that assigns entire precincts to Bush or Kerry makes little sense. More generally, any model that doesn't distinguish between individual respondents and precincts is going to confuse important issues, because both respondents and precincts matter.

[Time for change]

74. I can't speak for the others

I speculated that they would be in a better position to respond than I would, since I have very little experience with simulations and modeling. That doesn't necessarily mean that they would understand that particular post. In any event, two of these people have responded, and apparently they don't understand that particular post very well either.

[autorank]

75. First you speak for them, now you don't?

[Time for change]

76. I was just speculating about them, not speaking for them

[autorank]

77. Be honest and open, proud of your affiliations. "The truth shall set you

free."

[Bruce ODell]

57. USCV's Working Paper - and this analysis of WPE - is fatally flawed

I'm Bruce O'Dell - the Vice President and co-founder of US Count Votes.

With all due respect, I believe Ron's interpretation of Mitofsky's findings is fundamentally mistaken, and so is the USCV Working Paper, first published May 12.

After unsuccessfully working within US Count Votes to revise or retract the Working Paper that a minority of the USCV membership recently published, I see no alternative but to publicly challenge the report’s methodology and conclusions.

The key argument of the USCV Working Paper is that Edison/Mitofsky’s exit poll data cannot be explained without either (1) highly improbable patterns of exit poll participation between Kerry and Bush supporters that vary significantly depending on the partisanship of the precinct in a way that is impossible to explain, or (2) vote fraud. Since they rule out the first explanation, the authors of the Working Paper believe they have made the case that widespread vote fraud must have actually occurred.

However, a closer look at the data they cite in their report reveals that Kerry and Bush supporter exit poll response rates actually did not vary significantly by precinct partisanship. Systematic exit poll bias cannot be ruled out as an explanation of the 2004 Presidential exit poll discrepancy – nor can widespread vote count corruption. The case for fraud is still unproven, and I believe will never be able to be proven through exit poll analysis alone.

The fact that I chose not to endorse the USCV Working Paper should be a clear indication that I do not support its central thesis, and in fact believe that the simulation data they cite refutes the Working Paper’s conclusions.

I am not a statistician, but as a computer systems architect, I create mathematical models to simulate the performance of large-scale computer systems, and mathematical simulation of the cost and efficiency of business processes is a significant part of my consulting practice. My own election simulation results are cited on pp. 9 -10 and in Appendix G of the May 12th Working Paper; as the creator of the only USCV simulation which accurately reproduces aggregate Mean WPE, Median WPE and participation rate data from the E/M January report, I feel an obligation to ensure that my work is correctly interpreted.

I can show that several of the USCV election simulation programs are flawed, and that when the Liddle Bias Index is applied to the “USCV O’Dell simulation” data cited in the Working Paper, it produces results consistent with those recently reported by Warren Mitofsky for the E/M data as a whole.

I respect Ron's opinion, but his insistence on using aggregate WPE as a tool to interpret poll response bias (or vote fraud) is mistaken. His analysis of the Liddle Bias Index is also off-target. Liddle's Bias Index is an inherently superior metric to WPE, and analyses based on aggregate WPE are highly misleading.

I've written a paper that addressed this issues in detail, that can be found at www.digitalagility.com/data/ODell_Response_to_USCV_Working_Paper.pdf .

If anyone can show me where I'm wrong, I'll be the first to admit it.

I'm disappointed that I was not able to resolve our disagreement within USCV, but I simply cannot allow a fundamental misinterpretation of my data - the USCV O'Dell simulator they cite in their paper - to continue to go unchallenged.

In addition responding to this posting, please feel free to contact me at my email address at USCV, [email protected] - or at my corporate email address at [email protected] if you have any questions.

[Time for change]

58. Thank you for your clarification of this

In my response above (post # 56) I make some arguments that I believe are different than Ron's arguments that you are responding to here, but come to somewhat similar conclusions, and I would be very interested in your reaction to them. I do not disagree that the Liddle bias index is superior to WPE as a measure of bias, and I don't believe that any of my arguments contradict that.

[TruthIsAll]

66. Bruce, care to comment?

Bruce,

This optimization model determined that the maximum Bush% was
49.27% subject to the following constraints:

1) 53% weighted average response rate 
2) 50-56% Response range across all partisanship groups
3) 1.12 Alpha
4) The Kerry win % (shown below)

The maximum for Kerry, using the same criteria was 50.83%.

Since you have had extensive experience in modeling, I would
appreciate your comments.

The model (shown below in its entirety) is Excel-based and
uses the Solver optimization algorithm add-in. 

					Response Range	

	Kerry	50.73%	AvgResp	0.53	Low 	0.5
	Bush	49.27%	Alpha	1.12	High	0.56
						
		High Bush		High Kerry
	1250	40	415	540	165	90
	Kerry%	10%	30%	50%	70%	90%
						
53.0%	Resp	56%	56%	52%	50%	50%
						
663	Resp	22	232	280	82	45
349	K Resp 	3	79	157	65	45
314	B Resp	20	153	123	18	0
						
286	K Ref	2	55	130	58	41
302	B Ref 	16	127	130	25	4
						
634	TotalK 	4	134	287	123	86
616	TotalB 	36	281	253	42	4
						
	B Pct%	89.3%	67.6%	46.9%	25.6%	4.5%
	B Exit%	88.8%	66.0%	44.0%	21.4%	-0.9%
	B %Ref	90%	70%	50%	30%	10%
	WPE	0.5%	1.6%	2.9%	4.2%	5.4%



Here we assume an alpha of 1.0
Bush is a winner, which of course you would expect considering
that the partisanship groupings are weighted in favor of Bush.
But look at the response rates.

Bush precincts (45%); Kerry (65%)

					Response Range	
	Kerry	47.75%	AvgResp	0.53	Low 	0.45
	Bush	52.25%	Alpha	1	High	0.65
						
		High Bush			High Kerry
	1250	40	415	540	165	90
	Kerry%	10%	30%	50%	70%	90%
						
53.0%	Resp%	45%	45%	54%	65%	65%
						
662	Resp	18	187	292	107	59
332	K Resp 	2	57	146	75	53
330	B Resp	16	130	146	32	6
						
264	K Ref	2	69	124	41	28
323	B Ref 	20	159	124	17	3
						
597	TotalK 	4	126	270	116	81
653	TotalB 	36	289	270	49	9
						
	B Pct%	90.0%	69.6%	50.0%	29.8%	9.9%
	B Exit%	90.0%	69.6%	50.0%	29.8%	9.9%
	B %Ref	90%	70%	50%	30%	10%
	WPE	0.0%	0.0%	0.0%	0.0%	0.0%

[Bruce ODell]

68. I don't understand how to interpret your model

I'm sorry, but I don't quite see how to interpret your calculations.

I prefer to simulate exit poll participation on the precinct level, and then roll simulated precinct-level data up by category of partisanship to calibrate against E/M's aggregate numbers.

If you look at E/M's scaterplot, you see that there was tremendous variance in the precinct level statistics. Some precincts had WPE of about -60% (implying an extremely high Kerry/low Bush poll response rates, or equally well a huge vote shift to Bush). Some precincts had WPE of about +50% (implying an extremely low Kerry/high Bush poll response rate or equally well a huge vote shift to Kerry).

My point is that working with aggregate statistics greatly limits the behaviors that can be simulated, and working with aggregate WPE in particular inherently leads to skewed results - as I stated in my paper.

[Time for change]

67. I have started a new thread on Bruce O'Dell's statement

So if you want to respond to respond to his statement I suggest doing so on this thread: http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x374482

[Name removed]

80. Deleted message

Message removed by moderator. Click here to review the message board rules.

[Name removed]

84. Deleted message

Message removed by moderator. Click here to review the message board rules.

[RonB]

69. USCV's Analysis is not Based on WPE, "Aggregate" Analysis is not "Flawed"

Bruce,

You say:

"The key argument of the USCV Working Paper is that Edison/Mitofsky’s exit poll data cannot be explained without either (1) highly improbable patterns of exit poll participation between Kerry and Bush supporters that vary significantly depending on the partisanship of the precinct in a way that is impossible to explain, or (2) vote fraud. Since they rule out the first explanation, the authors of the Working Paper believe they have made the case that widespread vote fraud must have actually occurred."

I believe what we say is that varying patterns of partisan exit poll response require "further explanation" beyond a "pervasive bias" theory that is somehow inherent to being a Bush voter in that it is independent of precinct partisanship - our interpretation at the time of "rBr". The effort as I see it is to move the analysis beyond these vague and general "explanations" toward more serious substantive explanations that relate to the actual factors that influence response rate.

I don't believe we ever state that implausible invariant mean bias (K/B) is sufficient evidence of vote fraud. Just that a more substantive (so far not offered) explanation is required.

However, a closer look at the data they cite in their report reveals that Kerry and Bush supporter exit poll response rates actually did not vary significantly by precinct partisanship.

My analysis (for calculations see previous post - summary table copied below) demonstrates conclusively (I think) that K/B levels are indeed different at a 5% two-tail level of statistical significance for certain partisan categories of precincts. As you can see, it is particularly hard to see how the K levels calculated from both mean, and median (to eliminate out lier influence) WPE's for high Bush precincts could be obtained in any of the other categories of precincts.

These 95% confidence intervals are based on random sample exit poll selection - the same methodology use to analyze exit polls - without any cluster adjustment as there is no bias in estimating single precincts - and with mean approximates of R and b and k levels. If these later are not significantly correlated with each other (and the data suggest that they are not) these estimates should be quite accurate.

Means and Medians - and not individual data points, are after all, what statistical inference is usually about. Simulation in many ways is a "backward" reconstruction exercise, that cannot really add much (except for the additional hypotheses used to generate the simulations) to the original mean and median "information" about the data.


Table 2: Mean Based Calculations:

95% Confidence Interval for B 95% Confidence Interval for K
Covers all B? -1.96 x SD' B +1.96 x SD' Covers all K? -1.96 x SD' K +1.96 x SD'
Yes 20.9% 53.8% 86.6% No 49.3% 52.9% 56.6%
Yes 36.8% 49.6% 62.4% No 51.8% 57.3% 62.8%
No 43.2% 47.6% 52.0% No 52.0% 56.4% 60.8%
Yes 49.1% 52.6% 56.1% No 52.5% 60.6% 68.7%
Yes 47.1% 52.9% 58.7% Yes 31.9% 84.0% 136.1%


Table 4: Median Based Calculations:

95% Confidence Interval for B 95% Confidence Interval for K
Covers all B? -1.96 x SD' B +1.96 x SD' Covers all K? -1.96 x SD' K +1.96 x SD'
Yes 19.1% 51.9% 84.8% No 49.5% 53.1% 56.8%
Yes 37.1% 50.0% 62.8% No 51.7% 57.2% 62.7%
No 43.3% 47.7% 52.1% No 51.9% 56.3% 60.7%
Yes 49.1% 52.6% 56.1% No 52.5% 60.6% 68.7%
Yes 48.4% 54.2% 60.0% Yes 20.2% 72.2% 124.3%



You say:

"Systematic exit poll bias cannot be ruled out as an explanation of the 2004 Presidential exit poll discrepancy – nor can widespread vote count corruption. The case for fraud is still unproven, and I believe will never be able to be pr oven through exit poll analysis alone."

I agree.


You say:

"The fact that I chose not to endorse the USCV Working Paper should be a clear indication that I do not support its central thesis, and in fact believe that the simulation data they cite refutes the Working Paper’s conclusions.

I am not a statistician, but as a computer systems architect, I create mathematical models to simulate the performance of large-scale computer systems, and mathematical simulation of the cost and efficiency of business processes is a significant part of my consulting practice. My own election simulation results are cited on pp. 9 -10 and in Appendix G of the May 12th Working Paper; as the creator of the only USCV simulation which accurately reproduces aggregate Mean WPE, Median WPE and participation rate data from the E/M January report, I feel an obligation to ensure that my work is correctly interpreted."

Yes, definitely. The O'Dell simulation is an "output" simulation whose purpose is to simulate the E-M outcomes. The Dopp simulation is an "input" simulation whose purpose was to simulate the E-M hypothesis (or "hypothetical") inputs.


You say:

"I can show that several of the USCV election simulation programs are flawed, and that when the Liddle Bias Index is applied to the “USCV O’Dell simulation” data cited in the Working Paper, it produces results consistent with those recently reported by Warren Mitofsky for the E/M data as a whole."

I don't know what you mean by "flawed" but if you mean they don't reproduce the E-M outcomes thats the point! (see previous comment).

The Dopp simulator, in particular, as interpreted in Appendix I shows that the E-M outcomes are not obtainable under (what we then thought was) the E-M constant invariant K=.56, B=.5 hypothesis.


You say:

"I respect Ron's opinion, but his insistence on using aggregate WPE as a tool to interpret poll response bias (or vote fraud) is mistaken."

I think you're over stating your case here. There is nothing "inherently wrong" with "aggregate analysis", i.e. using means and medians. All these different types of analysis serve different purposes. Simulation adds the "information" (beyond the means and medians) that it assumes about the distributions to generate the simulations (see previous comments). This can be useful if there is some plausible "additional" information about these distributions and if there is a concern that "aggregation bias" resulting from parameters being correlated with each other (see previous post) will affect the analysis.

Using aggregate statistical analysis (as I do in the Tables above) is particularly useful to get significant probabilities, something that is particularly hard with simulations as there are so many possible options for distributions etc.


You say:

"His analysis of the Liddle Bias Index is also off-target. Liddle's Bias Index is an inherently superior metric to WPE, and analysis based on aggregate WPE are highly misleading."

I've stated all along that Liddle made an important contribution (that we acknowledged in the report) by focusing on looking at the "overall pattern" of K/B over partisan precincts. USCV had previously focused on calculating K and B in precinct quintiles and looking at the plausibility of these values and their ratios and differences across precinct quintiles, rather than at the overall pattern across precinct partisanship.

However, the basic point is that the calculation of and focus on K and B (the "confounded partisan response rates) was initiated by USCV. Liddle's alpha is simply LN(K/B). She derived it without K and B, apparently independently - its not a big deal to derive any of these formulas - the insight is in the need to do it and how they can be used. Liddle's contribution is in the later.


Unfortunately Liddle chose to frame her analysis as a new breakthrough in using "confounded variables" rather than simply a new way to use K and B to do the analysis. If she had been more clear about where her ideas came from, or at least who had done them before, she would have had to explain how she came to opposite conclusions from the same variables. I don't believe she did any of this deliberately but it has been the source of endless confusion that I think, most unfortunately, has gotten you confused as well.

NO BODY IS BASING ANY ANALYSIS ON WPE. ALL OF US ARE USING K AND B. WPE IS NECESSARY TO DERIVE K AND B BUT THE ANALYSIS IS BASED ON K AND B VALUES NOT ON WPE. WE ARE ALL ANALYZING THE SAME VARIABLES IN SLIGHTLY DIFFERENT FORMS (K, B, K/B, K-B. LN(K/B))

Liddle's other contribution was to point us in the direction of simulations. This is another method of analysis that could show the impact of "aggregation" effects. We used this in the report to show that these aggregation effects to not vitiate the conclusions of the aggregate analysis.


Ultimately this debate cannot be resolved with absolute certainty unless Mitofsky gives us the actual data (or does the calculations form the actual data himself). However, short of this I see no reason to make claims that simulations constitute absolute proof and "aggregate" statistical analysis is "off target" or fundamentally wrong.

The aggregate results show, with high probability, that K/B is not constant across precinct category. Your simulations show that, I guess, that this might not be the case. However, I don't think you (like Liddle) should jump to the conclusion that K/B is unvarying. It seems to me that our report (and the analysis above) makes a very plausible case that "more stuff needs to be explained" - and this is the point of it all!


Best,

Ron

[Peace Patriot]

78. Ron, you're right that "more stuff needs to be explained," like...

...why the news monopolies didn't commission exit polls specifically designed to verify this election--in view of the 2000 election fraud, and the many expert warnings about the hackability and insecurity of the new electronic voting systems being tested out nationwide for the first time?

...why the news monopolies ALTERED the results of the exit polls (Kerry won) to fit the official results (Bush won) on everybody's TV screens on election night, thus denying the American people major evidence of fraud?

...why the news monopolies and their pollsters are STILL withholding this evidence?

...why the news monopolies failed to report on the massive Voting Rights Act violations in Ohio and Florida, and John Conyers' investigation?

...why Tom Delay prevented a paper trail for electronic voting from ever getting out of committee in Congress?

...why Congress placed no controls on partisan ownership and control of electronic voting systems, and permitted SECRET, PROPRIETARY programming code in electronic voting machines and tabulators?

--why opinion polls show huge numbers of Americans disapproving of every major Bush policy, foreign and domestic, up in the 60% to 70% range, and dismal approval ratings for Bush (now at 42%), without any comment from the news monopolies or questioning of the "election," and, on the contrary, persistent efforts to create the ILLUSION that Bush and his fascist coup somehow have a "mandate"?

--why there has been not a single question by the news monopolies about the fact that Democrats blew the Bushites away in new voter registration in 2004 (nearly 60/40), with most new voters, most independent voter and most Nader voters voting for Kerry? (Where did Bush's margin come from?)

It seems pretty obvious that it is by design that we have this murky situation with the exit polls, as with the election itself.

[bleever]

81. Thank you for saying this so well, Peace Patriot. Science (mathematics,

in this case), is an instrument we turn onto phenomena in order to understand them, in their minutest aspects, and in their totality. Serious science, with all its necessary internal conflicts, choices, and controversies, is going on here.

But Science, in all its infinitely variable methods, mental models, and paradigms, doesn't choose where it will be focused. The decision on where to point the instrument arises in the social context.

And in the social context, in the national conversation about the 2004 election, there is tremendous gut reaction. Before we ever get around to directing the delicate instruments of analysis to dissecting the data, the human element finds many very well-informed people whose common sense knows that the results of this election are highly improbable, based on all the things you mention.

We have a brain with an analytic side, and an intuitive side. That's what makes a whole person. Analysis doesn't necessarily trump intuition; intuition isn't free from being held up to analysis.

All the analytical side can do is determine whether or not the final and absolute proof of an hypothesis has been found. The intuitive side determines whether or not to keep looking.


We'll keep looking. And not forget why. And at the same time, while still looking for the convicting piece of evidence, we don't need to waver in our confidence. We know what we know.


:thumbsup:

[RonB]

83. High Bush Precinct findings + Detailed Analysis of O'Dell Simulation Data

These have all been posted on the O'Dell discussion thread started by TruthIsAll.